Post on 28-Aug-2018
Mathematical Sciences School
Queensland University of Technology
Mathematical modelling of
tumour growth and interaction
with host tissue and theimmune system
Trisilowati
A thesis submitted for the degree of Doctor of Philosophy in the Science and
Engineering Faculty, Queensland University of Technology according to QUT
requirements.
Principal supervisor: Associate Professor Daniel Mallet
Associate supervisor: Associate Professor Scott McCue
2012
Keywords
tumour, immune, differential equation, cellular automata, optimal control, forward-
backward sweep, Runge-Kutta, dendritic cell, natural killer cell, cytotoxic T cell,
lymphocyte, helper T cell, CD4+, CD8+
i
Abstract
There is strong evidence in the literature for the hypothesis that tumour growth is
directly influenced by the cellular immune system of the human host. For example,
NK cells and CD8+ T cells are components of the immune system that are well
known to be able to kill tumour cells. Dendritic cells (DCs) as antigen-presenting
cells (APCs), play an important role in stimulating, recruiting and activating the
immune system. In recent years, it has been reported that DCs can also directly
lyse tumour cells. Some tumours present DCs and the presence of such cells has
a potential role in tumour control. Currently, mechanisms involved in immune
system interactions with growing tumours are not fully understood.
There are still many unanswered questions related to how the interaction be-
tween a growing tumour and immune system and regarding which components
of the immune system play important roles in this mechanism. The mathemati-
cal models to be developed in this project will provide theoretical descriptions of
biological systems that can be used to arrive at quantitative and qualitative un-
derstandings regarding answers to such questions as well as providing the ability
to simulate experiments in silico using computers. These in silico experiments
will help to extend current understanding and opinion related to tumour-immune
system interactions and allow for the development of more refined hypotheses that
can be tested via laboratory experiments.
In this thesis, three new mathematical models describing the growth of solid
tumours incorporating the host tissue and immune system response are developed
and investigated. We describe various biological aspects of tumour growth and im-
mune system interactions using mathematical models based on findings extracted
from the biological literature.
In the first investigation, two submodels are constructed to provide a description
of the interaction between a growing tumour and cells of the innate and specific
immune system. In these models, we assume that NK and DCs, (the innate immune
system), and CD8+ T cells (the specific immune system) can kill tumour cells. To
describe this interaction, the models are comprised of four ordinary differential
ii
iii
equations. We analyse the stability of the model as well as the simple bifurcation
behaviour. Numerical solutions of the models are also presented and interpreted.
A preliminary model is constructed to serve as an introductory description of
the biological system, before a more complex mathematical model is provided
which builds on the preliminary model with greater biological realism. Both of
the models demonstrate the compound effects of elements of the immune system
on the dynamics of tumour growth through their effects on other immune system
components. From numerical solutions, it is found that increasing the source term
of DCs is more useful than increasing the source term of NK cells.
In the second investigation, we study an optimal control treatment strategy
for a model of the interactions between a growing tumour and the host immune
system. To obtain the optimal control model, we extend the model of the first
investigation (described above) by introducing a time-varying dendritic cell-based
treatment strategy and also describing an objective functional that we seek to
minimise in this modelling study. Further, a discussion of a necessary condition for
an optimal strategy is presented to set the background of the model. The existence
of the optimal control is then proved. This then allows for a presentation of the
optimality system that will be solved numerically. The numerical scheme itself, a
forward-backward sweep method, is then applied to the model and a number of
important numerical solutions of the optimal control model are presented. From
simulations, increasing the strength of the DCV reduces the tumour burden and
the required time duration of the treatment. This model suggests that the best way
to fight the tumour is to give the first high DCV concentration at the beginning
of the treatment and reduce the treatment over the remaining treatment period.
In the final part of the thesis, we present a mathematical model of a growing
tumour and the interaction between the tumour cells and the host immune system
using a hybrid cellular automata (HCA) model. This model can describe the sys-
tem in much more detail, including cell-cell interactions of every single cell in the
system. To include the effect of a chemokine in this model, we recognise the signifi-
cantly smaller size of such molecules compared with biological cells and introduce a
partial differential equation to describe the concentration of chemokine secreted by
the tumour. We combine the numerical solution of the partial differential equation
with a number of biologically motivated automata rules to govern the evolution of
various cell populations to form the HCA model. From numerical simulations, the
tumour evolution shows the characteristic exponential and linear growth phases
of solid, avascular tumours (see for example, Folkman and Hochberg), as well as
a slower growing population of necrotic cells. The distribution of the growing tu-
mour is demonstrated with results qualitatively matching those of existing work
due to Mallet and de Pillis.
iv
We also use the HCA model to investigate the growth of a tumour in a number of
computational “cancer patients”. We present the results of these simulations using
a simulated Kaplan-Meier survival curve. From simulation, the model indicates
that increasing the number of immature DC in the domain results in significantly
longer “survival” of simulated patients.
Contents
1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 A review of relevant biological literature . . . . . . . . . . . . . . . 3
1.2.1 Tumours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 The immune system . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 The mathematical approach . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 Tumour growth models . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Modelling of cell migration . . . . . . . . . . . . . . . . . . . 13
1.3.3 Modelling tumours and the immune system . . . . . . . . . 20
1.3.4 Optimal control applied to tumour-immune system model . 29
1.3.5 Modelling tumours and the immune system using cellular
automata . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.4 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2 ODE models of specific immune cell interactions with a population
of tumour cells 39
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 Initial mathematical model . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.1 Non-dimensionalisation . . . . . . . . . . . . . . . . . . . . . 43
2.2.2 Steady state and stability analysis . . . . . . . . . . . . . . . 44
2.2.3 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3 A model incorporating Michaelis-Menten dynamics . . . . . . . . . 50
2.3.1 Steady state and stability analysis . . . . . . . . . . . . . . . 57
2.3.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
v
vi
3 An optimal control model of dendritic cell treatment of a growing
tumour 69
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Optimal control problem . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.1 Derivation of necessary condition . . . . . . . . . . . . . . . 72
3.2.2 Existence of an optimal control . . . . . . . . . . . . . . . . 75
3.2.3 Optimality system . . . . . . . . . . . . . . . . . . . . . . . 78
3.3 Numerical solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.1 Forward-backward sweep method . . . . . . . . . . . . . . . 81
3.3.2 Simulation and results . . . . . . . . . . . . . . . . . . . . . 84
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4 A 2D CA model of immune-tumour cell interactions 97
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.2 The role of DCs and chemokines . . . . . . . . . . . . . . . . . . . . 99
4.3 Mathematical model . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3.1 Diffusion equation for chemokine concentration . . . . . . . 104
4.3.2 Cellular automata rules . . . . . . . . . . . . . . . . . . . . . 104
4.4 Simulation and Results . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5 Conclusions and Discussion 130
5.1 Summary of thesis aims . . . . . . . . . . . . . . . . . . . . . . . . 130
5.2 Contribution of this thesis . . . . . . . . . . . . . . . . . . . . . . . 133
5.3 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
List of Figures
1.1 Schematic showing the humoral and cellular components of the im-
mune system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 The dimensionless tumour cell density resulting from the numerical
solution of the Perumpanani et al. model given by equations (1.3)
using MATLAB PDE solver pdepe for t = 0, . . . , 30, x = 0, . . . , 20,
n = 100 spatial points (fine mesh, short domain) and δt = 0.001. . . 16
1.3 The dimensionless tumour cell density resulting from the numerical
solution of the Perumpanani et al. model given by equations (1.3)
using MATLAB PDE solver pdepe for t = 0, . . . , 30, x = 0, . . . , 40,
n = 40 spatial points (coarse mesh, long domain) and δt = 0.001. . 17
1.4 The dimensionless tumour cell density resulting from the numerical
solution of the Perumpanani et al. model given by equations (1.3)
using MATLAB PDE solver pdepe for t = 0, . . . , 30, x = 0, . . . , 40,
n = 100 spatial points (fine mesh, long domain) and δt = 0.001. . . 18
1.5 Phase portrait of the Novozhilov et al. model given by equations (1.7)–
(1.8) using parameter values β = 1.5, γ = 0.7, and δ = 1. A number
of sample trajectories in the uninfected tumour cell–infected tumour
cell phase space are presented and we observe for this parameter set
the stability of the trivial steady state, reflecting clearance of the
tumour. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1 The role of DCs in killing tumour cells (reproduced from [81]). . . . 40
2.2 The evolution of tumour cells, NK cells, DCs and CD8+ T cells using
parameter values from Table 2.1 with the model given by equations
(2.1)–(2.4), showing a tumour cell population which grows initially
to a peak prior to full immune clearance. . . . . . . . . . . . . . . . 48
vii
viii
2.3 The evolution of tumour cells, NK cells, DCs and CD8+ T cells
using parameter values from Table 2.1 except for c1 = 3.5 × 10−7,
with the model given by equations (2.1)–(2.4), showing a tumour
cell population which grows initially to a peak prior to an oscilla-
tory response to the immune system and reaching a final, non-zero
equilibrium. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4 The evolution of NK cells showing the effect of varying the source
term of DCs in numerical solutions of the model given by equations
(2.1)–(2.4). Increasing the source term of DCs increases the peak
NK cell population due to the NK cell recruitment role played by
DCs and leads to more effective tumour cell clearance. Other pa-
rameters for these simulation taken from Table 2.1, with values of
s2 as indicated on the graph. . . . . . . . . . . . . . . . . . . . . . . 51
2.5 The evolution of CD8+ T cells in numerical solutions of the model
given by equations (2.1)–(2.4), showing that increasing the source
term of DCs increases the peak CD8+ T cell population again due
to the role played by DC-CD8+ T cell interactions in activating the
T cells. The effect here is more pronounced than it was for NK cells.
Other parameters for these simulation taken from Table 2.1, with
values of s2 as indicated on the graph. . . . . . . . . . . . . . . . . 52
2.6 The evolution of tumour cells in numerical solutions of the model
given by equations (2.1)–(2.4), showing the effect of varying the
source term of DCs. Increasing the source term of DCs decreases
the peak tumour cell population not only due to an increased DC
population but also the increases in NK and CD8+ T cells observed
in Figures 2.4 and 2.5. Other parameters for these simulation taken
from Table 2.1, with values of s2 as indicated on the graph. . . . . . 53
2.7 The evolution of tumour cells in numerical solutions of the model
given by equations (2.1)–(2.4), showing the effect of varying the
rate at which CD8+ T cell kill DCs, f1, with s1 = 5.000. Other
parameters for these simulation taken from Table 2.1, with values
of f1 as indicated on the graph. . . . . . . . . . . . . . . . . . . . . 54
2.8 The evolution of NK cells of the model given by equations (2.1)–
(2.4) show the population of NK cell growing in an uncontrolled
manner. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
ix
2.9 The evolution of tumour cells, NK cells, DCs and CD8+ T cells
found by numerically solving the model given by equations (2.10)–
(2.13) using parameter values from Tables 2.1 and 2.2, showing a
tumour cell population which grows initially to a peak prior to full
clearance by the combined immune system. . . . . . . . . . . . . . . 60
2.10 In the numerical solution of the model given by equations (2.10)–
(2.13), using parameters from Table 2.1 and 2.2, except for c1, bi-
furcation occurs at c1 = 1.62509 × 10−6. Shown here is that below
the bifurcation value, for example c1 = 1.6250 × 10−6, one tumour
cell, T (0) = 1, grows to the nonzero tumour equilibrium. . . . . . . 61
2.11 In the numerical solution of the model given by equations (2.10)–
(2.13), using parameters from Table 2.1 and 2.2, except for c1,
bifurcation occurs at c1 = 1.62509 × 10−6. In contrast to Fig-
ure 2.10, shown here is that above the bifurcation value, for example
c1 = 1.6251×10−6, an initial population T (0) = 1 grows but is then
controlled by the immune system, returning to a stable zero tumour
equilibrium. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.12 Numerical solution of equation (2.10)–(2.13) using data from Ta-
ble 2.1 and 2.2, except for c1 = 3.5 × 10−7 and l = 0.1. Here the
tumour grows to a high peak population initially before the immune
system is primed. The immune system is only able to reduce the
tumour population rather than clear it completely. . . . . . . . . . . 63
2.13 The evolution of tumour cells given by solutions of equations (2.10)–
(2.13) using parameters from Table 2.1 and 2.2, with different initial
values (indicated on the plot). This graph indicates that there is
a critical initial value of the tumour cell population, where all nu-
merical solutions of the model with initial value below the critical
value grow to the stable zero tumour equilibrium. However, all solu-
tions with initial value above the critical value grow to the nonzero
equilibrium point . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.14 The evolution of (a) tumour cells, (b) CD8+ T cells and (c) NK cells
showing the effect of varying the DC source term. Other parameters
for these simulations are taken from Table 2.1 and Table 2.2, with
s1 = 7.000 and values of s2 as indicated on the graph. . . . . . . . . 66
2.15 The evolution of (a) tumour cells, (b) CD8+ T cells and (c) DCs
showing the effect of varying the NKs source term. Other param-
eters for these simulations are taken from Table 2.1 and 2.2, with
s2 = 7.000 and values of s1 as indicated on the graph. . . . . . . . . 67
x
3.1 The evolution over time of the tumour cell population calculated via
the numerical solution of system (3.5)–(3.8) using different strengths
for the DCV (legend shown on the graph), administered to reduce
tumour burden. This plot shows that increasing the strength of the
vaccine decreases the magnitude of the peak tumour cell population
and also the time taken to reach the peak. . . . . . . . . . . . . . . 84
3.2 The evolution of two tumour cell populations using a large initial
tumour cell population, showing the impact of the vaccine-based
control strategy. The dashed line shows the tumour population
growth resulting from no control, while the solid line (more easily
seen inset) shows the tumour cell population eradication resulting
from the optimal control vaccine strategy. . . . . . . . . . . . . . . 85
3.3 The optimal control, u, used for treatment of the tumour cell pop-
ulation shown in Figure 3.2. . . . . . . . . . . . . . . . . . . . . . . 86
3.4 The evolution of three NK cell populations showing the impact of
the DC vaccine-based control strategy for three different maximum
vaccine levels (as shown in the legend). This plot indicates that
increasing the maximum vaccine level increases the peak NK level
and decreases the time to reach that peak. . . . . . . . . . . . . . . 87
3.5 The evolution of three CD8+ T cell populations showing the impact
of the DC vaccine-based control strategy for three different max-
imum vaccine levels (as shown in the legend). This plot indicates
that increasing the maximum vaccine level increases the peak CD8+
T cell level and slightly decreases the time to reach that peak. . . . 88
3.6 The evolution of three tumour cell populations showing the impact
of the DC vaccine-based control strategy for three different maxi-
mum vaccine levels (as shown in the legend). This plot indicates
that increasing the maximum vaccine level decreases the peak tu-
mour cell level and slightly decreases the time to reach that peak. . 89
3.7 The control, u, used for treatment with initial value of tumour cells
T0 = 42310 and where the maximum vaccine to be administered is
(a) 2000 units and (b) 20000 units. . . . . . . . . . . . . . . . . . . 90
3.8 The control, u, used for treatment with initial value of tumour cells
T0 = 42310 and where the maximum vaccine to be administered is
200000 units. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
xi
3.9 The evolution of two tumour cell populations using T0 = 100. Pa-
rameter values taken from Tables 2.1 and 2.2, except for c1 =
1.5× 10−6. The dashed line shows the tumour population resulting
from no control, while the solid line (more easily seen in the inset
plot) shows the tumour cell population eradication resulting from
the optimal control vaccine strategy. . . . . . . . . . . . . . . . . . 92
3.10 The control, u, used for treatment with an initial value of tumour
cells T0 = 100 and where the maximum vaccine to be administered
is (a) 2000 units and (b) 20000 units. Parameter values are taken
from Tables 2.1 and 2.2, except for c1 = 1.5× 10−6. . . . . . . . . . 93
3.11 The control, u, used for treatment with an initial population of tu-
mour cells T0 = 100 and where the maximum vaccine to be admin-
istered is 200000 units. Parameter values are taken from Tables 2.1
and 2.2, except for c1 = 1.5× 10−6. . . . . . . . . . . . . . . . . . . 94
3.12 The evolution of three tumour cell populations using optimal control
with varying maximum maximum vaccine levels to be administered
(as shown on the figure legend). Parameter values are taken from
Table 2.1 and 2.2, except for c1 = 1.5×10−6. Intuitively, the plot in-
dicates that increasing the maximum value for the vaccine decreases
both the peak of tumour cell population and the length of time to
eliminate the tumour cells. . . . . . . . . . . . . . . . . . . . . . . . 95
4.1 Schematic showing the partitioning of the problem domain into cel-
lular automata elements (squares) and mesh-points for numerical
solution of the partial differential equation. . . . . . . . . . . . . . . 102
4.2 The form of the curves used to determine the probability of (a)
tumour cell division and (b) tumour cell lysis, given different neigh-
bourhood conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3 The growing tumour and host immune system. After 25 cell cycles
(a). After 50 cell cycles (b). After 75 cell cycles (c). After 100 cell
cycles (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4 Total cell counts of tumour and necrotic cells after 100 cell cycles.
This plot shows a slower growing population of necrotic cells and an
exponential growth of tumour cells . . . . . . . . . . . . . . . . . . 118
4.5 Total cell counts of tumour cells for 100 simulations (thin) and the
median simulation (thick) of CA model over 100 cell cycles. . . . . . 119
4.6 Total cell counts of tumour cells for 100 simulations showing the ef-
fect of varying the DCs as indicated on the graph over 100 cell cycles.
This plot shows that increasing the population of DCs decreases the
number of tumour cells. . . . . . . . . . . . . . . . . . . . . . . . . 120
xii
4.7 Total cell counts of CD8+ T cells (a) and DCs (b), after 100 cell
cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.8 Total cell counts of CD4+ helper T cells (a) and NK cells (b), after
100 cell cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.9 The evolution of CD8+ T cells, Dendritic cells, T Helper cells, NK
cells, healthy cells, tumour cells and necrotic cells for 5 simula-
tions over 300 cell cycles. Parameters used are: I0 = 0.005, D0 =
0.002, H0 = 0.002, K0 = 0.001, Pdiv = 0.5, Pmig = 0.2 . . . . . . . . . 123
4.10 The evolution of CD8+ T cells, Dendritic cells, T Helper cells, NK
cells, healthy cells, tumour cells and necrotic cells for 5 simula-
tions over 300 cell cycles. Parameters used are: I0 = 0.009, D0 =
0.002, H0 = 0.002, K0 = 0.001, Pdiv = 0.9, Pmig = 0.2 . . . . . . . . . 124
4.11 Simulated Kaplan-Meier curve with different initial values of imma-
ture DCs as indicated on the graph. This plot shows that increasing
the population percentage of DCs in patients will increase “survival”
rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.12 Simulated Kaplan-Meier curve with different initial values of imma-
ture CTL as indicated on the graph. This plot shows that increas-
ing the population percentage of CTL cells in patients will increase
“survival” rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
List of Tables
2.1 Parameter values and their associated sources used in numerical
solutions in the present research. . . . . . . . . . . . . . . . . . . . 47
2.2 Parameter values used in numerical solution of equations (2.10)–
(2.13) in addition to those in Table 2.1. . . . . . . . . . . . . . . . . 59
4.1 The different cell species tracked in the cellular automata and the
numerical value given to each in the computational implementation. 103
4.2 The variables used in the hybrid cellular automata model. Here x
and y are the spatial variables for the PDE component, t denotes the
time variable, and i and j represent spatial locations in the cellular
automata component. . . . . . . . . . . . . . . . . . . . . . . . . . . 105
xiii
Statement of Original Authorship
The work oatained in thi* thesig has not been previously ilbmitted for a degree
or diploma at a,ny other higler educational inetitution. To the besi of my kaowl-
edge an<t behef, the thesis sontnins no mataial prwiously published or vdtten byanother pereon except where due reference is made,
lqv
Acknowledgements
First and foremost, I offer my humble thanks to Allah for giving me the opportu-
nity, strength and guidance to undertake my PhD.
Secondly, I would like to sincerely express deep gratitude to my principal super-
visor, Associate Professor Daniel Mallet, for his valuable assistance, counselling,
support, patience and kindness. Without his help, I would not have been able to
finish my study. Words are not enough to express how grateful I am. I owe him
much.
My sincere thanks also go to my associate supervisor, Dr Scott McCue, for his
guidance, suggestions and help over the past three years.
I also thank my sponsor, Directorate General for Higher Education (DIKTI),
Ministry of Higher Education, Indonesia, for a scholarship for postgraduate study.
Most importantly, I would like to give my special thanks to my father for his
support, my husband for his patience and support and my children, whose love
and support have sustained me during my study.
Finally, I would like to thanks my friends in Room O 502, especially Masoum,
for their help and support.
xv
Chapter 1
Introduction
1.1 Overview
Cancer is one of the leading causes of death worldwide, with 7.6 million people
dying as a result of cancer in 2008 alone. This is projected to rise to over 13.1
million by 2030 [131]. A similar report [4] states that in Australia in 2007, cancer
was the second most common cause of death and that 108, 368 new cases of cancer
were diagnosed. For those diagnosed with cancer between 1998 and 2004, the 5-
year relative survival for all cancers combined was 61%. Clearly, cancer is a major
concern for public health officials around the world and a greater understanding
of cancer has potential to save many lives.
There is evidence that the immune system is capable of recognising and eliminat-
ing tumour cells [16, 31, 106, 122]. Therefore, some researchers intensively continue
to develop and investigate theoretical and experimental approaches regarding the
interactions between growing tumours and the immune system. However, it is
difficult to observe and to control experimentally all of the interacting elements
of a growing tumour due to the sophistication of the biological system which de-
pends on many factors involved in the process [92]. Mathematical models play an
important role in the development of knowledge in this field of research, since we
can use models to understand the general behaviour of a phenomenon in different
situations, to perform in silico experiments or simulations, to carry out new exper-
iments, and to test theoretical assumptions and suggest modifications of theories
[85]. The development of mathematical models of tumour growth and immune re-
sponses requires knowledge in two different areas, in particular: the understanding
of biological phenomena involved in the growth and response processes, and also in
using a variety of mathematical tools to obtain both qualitative and quantitative
predictions [108].
There are still many unanswered questions related to how the interaction be-
1
2
tween a growing tumour and immune system and regarding which components of
the immune system play important roles in this mechanism [39]. The mathemati-
cal models to be developed in this project will provide theoretical descriptions of
biological systems that can be used to arrive at quantitative and qualitative un-
derstandings regarding answers to such questions as well as providing the ability
to simulate experiments in silico using computers. These in silico experiments
will help to extend current understanding and opinion related to tumour-immune
system interactions and allow for the development of more refined hypotheses that
can be tested via laboratory experiments. It is important to study or develop such
models, because these models can provide valuable input to the development and
refinement of tumour treatment strategies.
In this thesis, three new mathematical models describing the growth of solid
tumours incorporating the host tissue and immune system response will be devel-
oped and investigated. We attempt to describe the various biological aspects of
tumour growth and immune system interactions using mathematical models based
on findings extracted from the biological literature. First, a differential equation
model will be constructed to investigate the relationship between tumour growth,
host tissue and immune system components. Then we will expand the model to
explore immunotherapeutic treatment in the form of dendritic cell application, and
the effects of such treatment on tumour growth. This will be explored using an
optimal control framework. We also investigate a hybrid cellular automata model
of tumour-immune system interactions to undertake an investigation that not only
allows for the investigation of spatial variation in tumour growth and immune
response, but also for variation across computational “cancer patients”.
There are three aims that will be carried out in this thesis. They are:
Aim 1: To develop a differential equation-based mathematical model of a growing
tumour, host tissue and immune system, the associated interactions and
resulting outcomes. Based on de Pillis and Radunskaya model [34] and
Castiglione and Piccoli [26], we construct new mathematical models that
explain more detail the role of DCs incorporating with NK cells, CD8+
T cells and tumour cells. These models can provide a tool to describe
qualitative relationships based on particular laboratory findings.
Aim 2: To develop a mathematical model of DC-based immunotherapeutic treat-
ment of a growing tumour, that can provide a basic level of understand-
ing and feedback to experimentalists to assist in designing more effective
and/or efficient laboratory experiments.
Aim 3: To develop a cellular automata model of tumour-immune system inter-
actions that explicitly accounts for cell-cell interactions, incorporates a
3
multi-dimensional spatial viewpoint and allows for stochastic variation to
simulate virtual patients. Based on Mallet and de Pillis model [89], we
build a new cellular automata model that describes the immune system in
more detail. The effect of chemokines in a growing tumour and the sim-
ulation of virtual patients using similar Kaplan-Meier curve also provide
a new contribution to the literature. This model then provides a deeper
level of understanding of the immune system interactions with a growing
tumour.
To this end, the research requires a working understanding of biological and math-
ematical literature and methods. The mathematical models developed incorporate
differential equations, partial differential equations, cellular automata and opti-
mal control theory and require the application of numerical schemes such as the
finite difference method, forward-backward sweep with Runge-Kutta differential
equation solvers, and purpose-written algorithms for cellular automata.
The remainder of this chapter is organised as follows. First a review of biological
literature of relevance to the problem of interest is presented. This includes a
discussion of the immune system, the growth of tumours and the interaction of
the immune system with a growing tumour. Then a brief review of some relevant
mathematical models is presented. This review includes models which simply look
at tumour growth alone, as well as modelling that incorporates elements of the
host immune response. Technical aspects are also reviewed and in particular, a
discussion is presented of the use of optimal control theory and of cellular automata
in modelling tumour growth. Finally, the structure of the thesis as a whole, as well
as its main findings are summarised.
1.2 A review of relevant biological literature
1.2.1 Tumours
A general discussion regarding what a tumour is can be found in many references.
For example, the Australian Institute of Health and Welfare discussion forms an
appropriate basis for this work and is summarised as follows [5]. Normally, cells
grow and replicate in an orderly way and the generation of tissue and organs
occurs to satisfy a particular function in the body. Uncommonly, however, after
being affected by a carcinogen, or after developing a random genetic mutation,
cells replicate in an uncontrolled way and form a mass which is called a tumour or
neoplasm.
Tumours can be classified into two groups, namely benign tumours that do not
spread to other tissue and malignant tumours that spread or invade to other tissue.
4
Benign tumours are often thought to be less dangerous than their invasive malig-
nant counterparts, however it should be noted that the growth of a benign tumour
can be dangerous due to the resulting obstruction of natural bodily functions. The
main characteristic of a malignant tumour is its ability to grow in an uncontrolled
way and to invade or spread to other tissues in the body. Tumour cells spread
to other parts of the body when the bloodstream or the lymphatic system carries
some cancer cells and lodges them some distance away. They can then start to
form a new tumour and continue to invade again. Some tumours can stay in the
body for years without showing any symptoms. Others can grow, invade or spread
quickly, and cause death in a short period of time.
Cell migration is an essential factor in tumour cell invasion. Migration can be
promoted by numerous factors including chemical substances, pressure gradients
and components of the extracellular matrix (ECM). Normally, cells in tissue are
attached to the ECM and also to one another in specific ways related to the
purpose of those cells. Interruption of this adhesion leads to increased motility
of cells and possible invasiveness of cells through the ECM. Cell surface receptors
that mediate cell adhesion are called integrins. The interactions between cells,
via such receptors, are also crucial in the normal regulation of proliferation and
differentiation of cells [117]. These interactions are mediated by the cell adhesion
molecules (CAMs) which is a family of molecules expressed at the surface of the
cell [112, 117].
1.2.2 The immune system
An immune system is a collection of biological mechanisms and processes inside
an organism with the purpose of protecting the organism against diseases and
infections by identifying and killing non-self (foreign) matter such as viral particles,
parasites, and which is of importance here, tumour cells. There is a number of
ways to provide a classification for the human immune system and one involves
splitting the system into two components, namely the innate and adaptive systems.
The innate immune system can recognise foreign matter without the requirement
for previous priming by specific non-self antigens. The adaptive immune system
can recognise specific targets, after antigens have been processed and presented
in combination with a self-receptor or major histocompatibility complex (MHC)
molecule on the surface of a cell, resulting in the production of memory immune
cells [117].
Innate immunity can involve phagocyte cells such as neutrophils and macrophages,
natural killer (NK) cells, dendritic cells (DCs), cytokines, and the complement
system. NK cells, DCs and macrophages are important in tumour recognition.
Numerous observations have indicated that activated macrophages also play a sig-
5
nificant role in the immune response to tumours [73]. In preventing the develop-
ment of clinical tumours, NK cells also play a key role by destroying abnormal cells
before they replicate and grow [39]. NK cells have the ability to recognise certain
tumour antigens and to kill the tumour cells [117]. In recent years, DCs have been
identified as an important component in controlling the growth of tumours. Also
DCs have a potential role in directly killing the tumour cells [81].
Dendritic cells are known as antigen-presenting cells (APCs), which have an im-
portant role in activating the immune system. DCs uptake and process the antigen
in peripheral tissue and then migrate to lymphoid tissue where antigen presenta-
tion to the immune system occurs [11, 123]. They present antigens to CD8+ T
cells through MHC class I molecules and they are very effective in activating CD4+
helper T cells through MHC class II molecules. Once activated, CD4+ helper T
cells then secrete chemokines that can enhance immunoglobulin production. Den-
dritic cells also play a vital role as the major regulator in cytotoxic T lymphocytes
(CTLs) and NK cell activation [30, 51, 81].
Adaptive immunity can involve antigen presenting cells, T and B lymphocytes,
cytokines and the MHC system. T cells and B cells are derived from the process
of hematopoiesis in the bone marrow. T cells are involved in the cell-mediated
immune response while B cells are involved in the humoral immune response (that
is, antibody-mediated). These two arms of the immune system are illustrated in
Figure 1.1. Furthermore, T cells can be split into two groups: the T helper cells
and the cytotoxic T cells. T Helper cells regulate to activate both the innate
and adaptive immune responses and can only recognise antigen that is presented
together with class II MHC on the cell surface aided by a coreceptor on the T cell,
called CD4+, whereas cytotoxic T cells generally can recognise antigen combined
with the MHC Class I aided by the CD8+ coreceptor [73]. On the other hand, the
B cell antigen-specific receptor is an antibody molecule on the B cell surface and
is presented with the MHC class II [117].
Tumour immunology involves the study of the host immune response to antigens
on tumour cells [73]. There are two types of tumour antigens that have been
identified on tumour cells: tumour-specific transplantation antigens (TSTAs) and
tumour-associated transplantation antigens (TATAs). Tumour-specific antigens
which are unique to tumour cells do not occur in normal cells. They may result
from a mutation in the tumour cells that can alter cellular proteins. They are
then presented with class I MHC molecules, invoking a cell-mediated response by
tumour-specific CTLs. However, tumour-associated antigens are not unique to
tumour cells which may be proteins. They are expressed on normal cells during
fetal development when the immune system is immature and unable to respond.
As a result of destruction of tumour cells, tumour antigens are produced that
6
AHumoral immune response Cellular immune response
+B lymphocyte Antigen
ASPC
Eliminating antigen
Dendritic cells
Act’d T helper Act’d cytotoxic
T helper eff Cytotoxic T eff
Cytokines Killing infected cell
Figure 1.1: Schematic showing the humoral and cellular components of the immune system.
influence both the humoral and cell-mediated immune responses [73]. In general,
the cell-mediated response appears to play the major role in a response to a grow-
ing tumour. A number of tumours have been shown to induce tumour-specific
CTLs that recognise tumour antigens presented by class I MHC on the tumour
cells. Research has shown that costimulatory signal required for activation of CTL
precursors can enhance tumour immunity. A variety of experimental and clini-
cal approaches have been developed to use recombinant cytokines to augment the
immune response against cancer either singly or in combination with each other.
Treatment
There are three traditional therapy procedures practised for treatment of tumours:
surgery, radiation therapy, and chemotherapy [98]. Surgery attempts to directly
remove tumours and hence reduces tumour burden. Radiation therapy destroys
and kills tumour cells by the direct application of radiation to the affected area.
Chemotherapy attempts to destroy tumour cells with drugs. However, all these
procedures are identified by a relatively low efficacy and high toxicity for the
patient. The prospect of treatment through immunotherapy is a relatively new
regarding tumour treatment and offers promise due to the fact that it mimics the
natural response to non-self entities.
Immunotherapy usually involves the use of cytokines together with adoptive cel-
lular immunotherapy (ACI). Cytokines are protein hormones that mediate both
natural and specific immunity [77, 98]. Among the cytokines that have been
evaluated in cancer immunotherapy are interferon-α (IFN-α), IFN-β, and IFN-
7
γ, interleukin-2 (IL-2), IL-4, IL-6, and IL-12; granulocyte-macrophage colony-
stimulating factor (GM-CSF) and tumour necrosis factor (TNF) [73]. The cy-
tokines are produced mainly by activated T cells (lymphocytes) during the cell-
mediated immune response. Interleukin-2 produced by CD4+ T cells is the main
cytokine responsible for lymphocyte activation, growth and differentiation. ACI
refers to the injection of cultured immune cells with anti-tumour reactivity, into a
tumour bearing host [77]. This type of treatment consists of two approaches:
• LAK (lymphocyte-activated killer cell) therapy: These cells are obtained
from the high concentration of IL-2 in peripheral blood leukocytes taken
from patients, via in vitro culturing. The LAKs are then injected back at
the tumour site.
• TIL (tumour infiltrating lymphocyte) therapy: These cells are obtained from
in vitro incubating with a high concentration of IL-2 lymphocytes recovered
from the tumour itself and are comprised of activated NK cells and CTL
cells. They are then injected back at the tumour site.
1.2.3 Summary
In this section we have presented a brief review of relevant biological literature.
In particular, tumours themselves have been discussed in terms of classification,
composition and means of affecting the host. The immune system has also been
discussed and in particular we have described different types of immune response as
well as different arms of the immune system and the associated cell types. Finally, a
short overview of different types of tumours treatment, including the relatively new
approach of immunotherapy, has been presented. This background information is
important as it is referred to, built upon and relied on in the construction of the
mathematical models presented later in this thesis. It is this information, along
with further detail presented in the subsequent chapters, which will inform the
biological relevance of the models.
1.3 The mathematical approach
In this thesis, two fundamentally different mathematical methodologies are em-
ployed in the modelling of tumour growth and subsequent immune system in-
teractions. In particular, first we use ordinary differential equations to form a
description of growth and interactions, that is then coupled to an optimal con-
trol strategy to investigate tumour treatment. Then we employ a hybrid cellular
automata model to undertake an expanded investigation of the system.
8
The ordinary differential equations (ODEs) modelling approach are extremely
well tested in this context (see the review presented in the coming sections), but
while it has several strong points as a modelling strategy it is not without its dis-
advantages. Ordinary differential equations, particularly those traditionally used
to model tumour growth (which are not terribly nonlinear) are fairly easily solved
in the numerical sense as well as in many cases being amenable to analytical inves-
tigations such as via phase plane, stability and bifurcation analysis. ODEs allow
the researcher to look at changes in the dynamics of the system in a sense that is
similar to how experimental researchers conduct their investigations – that is, the
natural output of a system of ODEs is the time course of each variable of inter-
est, just as experimentalists record observations of a system over time. However,
restricting a study to the use of ODEs imposes an assumption that the system is
spatially well-mixed. This is of course problematic when spatial variation is impor-
tant in a system. Furthermore, ODEs of the type used in most tumour modelling
work only allow for consideration of population-level interactions. When individual
interactions between cells and/or between cells and other matter are important,
then the usefulness of ODE models is decreased.
More recently, discrete approaches such as cellular automata (CA) and hybrid
cellular automata (HCA) models have been used for modelling tumour systems.
While simple conceptually, such models do not facilitate analytical investigations
except in the most basic (completely unrealistic) cases. However, CA models are
straightforward to implement computationally and the lack of a means of general
analysis is compensated by the ease with which the models can be simulated in
silico. Simple simulation also means that where system parameters are unknown, it
is possible to computationally investigate the possible parameter space. A further
advantage of CA models is that they do allow for individual level interactions to
be captured. In particular, each individual cell can be investigated for interactions
with other cells, other matter in the region, forces, chemical gradients and even
internal variation, at any point in the progression of the model solution.
A review of some of the relevant existing mathematical models of tumour systems
is now presented in the remainder of this section. In this review, some of the ideas
discussed above are elaborated upon in the context of the studies that have already
taken place. The review also demonstrates that the methodologies applied in the
current research are well-placed in terms of existing work in the field. Presented
first is a review of some of the earliest models of tumour growth based on nutrient
diffusion, as well as an inspection of some cell migration models. Then models that
incorporate the immune system and use an ODE approach are covered, including
ODE models of tumour growth and the immune system that are coupled with
optimal control analysis. Finally, we briefly discuss CA-based models of tumour-
9
immune system interactions.
1.3.1 Tumour growth models
The dynamics of tumour growth and the interactions of growing tumours with
the host immune system have been a significant focus for mathematical modelling
over the past four decades [89]. Araujo and McElwain [10] recently presented
an excellent review of the mathematical modelling of tumour growth. Starting
from early chemical diffusion and differential equation models of Burton [22] and
Greenspan [63, 64], tumour modelling has expanded to include ordinary differential
equations (ODE) models, partial differential equation (PDE) models [2, 14, 62, 89,
92, 107, 109, 128] and cellular automata (CA) models [89]. Increasingly, models
are being explicitly tied to experimental results and data, for example, in [74], a
mathematical model based on the diffusion of nutrient especially for oxygen and
glucose is developed and is validated using in vitro tumour growth data.
Multicell spheroids have been studied for some time now as laboratory models
for in vivo tumours and over the past few decades, they have become the focus of
modelling efforts of applied mathematicians. Multicellular spheroids can consist of
three distinct regions: a central necrotic core, an inner shell of quiescent cells and
an outer shell of proliferating or viable cells. Among the earliest models of tumour
spheroids was the work of Greenspan [63] who developed a simple model of growth
depending on nutrient diffusion. Since then, many others have modelled various
aspects of tumour spheroid growth including nutrient limited growth, immune
system interactions, effects of stresses and cell migration (see for example, Pettet
et al. [110]). Here we present a summary of the pioneering work of Greenspan to
set the scene for the modelling to be considered in subsequent sections.
Greenspan (1972)
Greenspan proposed a simple mathematical model of tumour growth by diffusion
in order to investigate the evolution of solid carcinoma [63]. The growth of a solid
tumour in the earliest stage is regulated by the direct diffusion of nutrient and
wastes, to and from surrounding tissue. By simple diffusion, each cell receives
adequate nourishment when the size of the tumour is very small and the resulting
growth rate of the population is exponential. There is a critical tumour size where
the growth rate of the tumour then diminishes markedly. Greenspan describes the
tumour shape as a sphere that consists of three layers: a central necrotic core, a
layer of viable non-proliferating cells and the outer shell where all mitosis occurs.
To model this tumour growth, the following assumptions are made.
1. The shape of a solid tumour is a sphere and complete spherical symmetry
10
prevails at all times.
2. When the concentration of a crucial nutrient falls below a critical level, tu-
mour cells die.
3. The vital nutrient is consumed by living cells only.
To describe the characteristics of a growing tumour and the dormant steady
state, there are some new hypotheses and approximations that must be added,
such as adhesion, disintegration of necrotic cellular debris, and the production
of a chemical that inhibits the mitosis of cancer cells without causing their death.
Finally, it is assumed that the carcinoma is in a steady state of diffusive equilibrium
at all times. To construct the model of tumour growth, a conservation of mass
principle is applied such that
A = B + C −D − E, (1.1)
where A represents the total volume of living cells at any time t, B is the initial
volume of living cells, C denotes the total volume of cells produced in t > 0, D
is the total volume of necrotic debris at time t and E represents the total volume
lost in the necrotic core in t > 0.
These terms can be written mathematically as
A =4π
3(R3
0(t)−R3i (t)),
B =4π
3(R3
0(0)),
C = 4π
∫ t
0
dt
∫ R0(t)
max(Ri(t),Rg(t))
S(σ, β)r2dr,
D =4π
3R3i (t),
E =4π
3
∫ t
0
3λR3i (t)dt,
where R0(t) is the outer radius of the tumour at any time t, Ri(t) is the radius
of necrotic core, Rg(t) is the radius at which cell proliferation ceases, S(σ, β) is
the proliferation rate of cells, σ and β represent the concentrations of nutrient and
inhibitor respectively, and 3λ is the proportionality constant for the rate at which
the necrotic core loses cell volume.
Substituting each of these expressions into equation (1.1) and differentiating the
resulting equation with respect to t gives the integro-differential form of the outer
boundary condition
R20
dR0
dt=
∫ R0
max(R1,Rg)
S(σ, β)r2dr − λR3i ,
11
which relates the tumour radius to the volume production due to mitosis and
volume loss due to necrosis.
In this paper, a model of growth retardation due to necrosis and wastes from
living cells is also derived. For the growth retardation due to necrosis model,
during the first phase, it is found that the tumour grows at an exponential rate
until the first cell at the centre of the sphere dies due to lack of sufficient nutrient.
In the second phase, the growth either tends to a final steady state or it terminates
at some point. Finally, in the third phase, a period of retarded tumour growth
occurs because of the death of cells and because chemical inhibition of mitosis is
achieved. There are also three phases of development in tumours which exhibit
growth retardation due to wastes from living cells. In the first phase, the tumour
grows at an exponential rate until growth retardation occurs at the centre of the
sphere. The growth rate changes from the initial exponential rate to one that is
approximately linear in time in the second phase. In the last phase, this model
indicates that the rate of volume loss per unit volume in the necrotic core controls
the tumour growth resulting in a steady state tumour volume.
Greenspan (1976)
In a subsequent paper, Greenspan explained the distribution of nutrients related
to the growth and movement of certain cell cultures and solid tumours [64]. He in-
vestigated the unstable development of tumours when the internal pressure forces
overcome surface tension and adhesion. The processes and mechanisms of cell
culture growth are very complex and in order to reproduce the main qualitative
features of shape and structure, Greenspan made a number of simplifying assump-
tions. Some of these have been discussed by Greenspan in [63] (see above). Others
include:
1. The culture and surrounding medium is essentially in diffusive equilibrium
at all times. It is assumed that the tumour has two-layers: an outer shell
and a larger core of necrotic debris.
2. When the available concentration of a vital nutrient, denoted by σ(x, y, z, t),
falls below a critical level σ1, cells begin to die. If h is the local thickness
of the thin layer of living cells and this depth depends only on σ1 and the
value of σ at the outer surface of tumour, this relationship can be written as
follows
h =
ν√σ − σ1, σ > σ1
0, σ < σ1.
3. The mitotic index is a constant.
12
4. Necrotic debris breaks down continually into simpler compounds.
5. The force of surface-tension Γ is proportional to the mean deflection κ of the
boundary.
6. Internal pressure differentials which cause the motion of cellular material are
produced by the birth or death of cells and it is assumed that
q = −∇p,
where q(x, y, z, t) is the particle velocity and p(x, y, z, t) is proportional to
the internal pressure.
Using these assumptions, the mathematical model of cell cultures and solid tu-
mours is formulated as
∇2p = Si inside Γ = 0,
∇2σ = 0 outside Γ = 0,
where Si is the rate of volume loss per unit volume, and on Γ(x, y, z, t) = 0 we
have
p =α
2(κ1 + κ2) = λκ,
q+ · n̂ = −n̂ · ∇p+ λ√σ − σ1,
q+ × n̂ = −∇p× n̂,
n̂∇σ = µ√σ − σ1.
The bounding surface is defined by
dr
dt= q+,
and the initial parametric prescription
r = a(ξ, ζ) at t = 0.
This model describes the relationship between the nutrient concentration, the pres-
sure on the surface and the surface-tension force. Greenspan states that if the tu-
mour reaches a critical size beyond which surface tension is overcome by pressure
forces, the tumour becomes unstable. It is shown that in the necrotic core, the
propensity of the colony to distort by either growth or the elimination of material
can reverse the effect of stability on the surface tension and on the other hand, by
controlling the distribution of nutrient, a steady state equilibrium can be reached.
The work of Greenspan is important here as it provides the basis for much of the
tumour modelling that followed. Ordinary and partial differential equation-based
mathematical models of tumour growth, nutrient supply, contaminant removal and
so on, provide a starting point for all of the mathematical models developed in this
thesis.
13
1.3.2 Modelling of cell migration
The cell migration-dependent invasion of tumour cell colonies is examined by a
number of authors (see for example [88, 107, 108, 109]). Mallet [88] for example,
explained the role of haptotaxis in tumour growth. Perumpanani [107] constructed
a mathematical model of tumour growth that considers the way in which chemo-
taxis and haptotaxis cooperate to regulate invasion, whereas in [109], Perumpanani
and coauthors developed and analysed a model for malignant invasion that com-
bined both proteolysis and haptotaxis. Integrins, as described in [90], also play
an important role in cell migration and recently integrins have become the topic
of some experimental investigations [113]. Below we discuss the background, con-
struction and analysis of some of these models.
Perumpanani et al. (1998)
During the process of migration, cells use a combination of changes in adhesion,
proteolysis and motility (directed and random). Haptotactic gradients, which cells
use to move in a directed fashion, are produced by proteolysis of the ECM. The
ability of the migratory cells to secrete enzymes and digest ECM as they move
through it, is a critical feature of cell migration. Invasive cells in vivo bind to sur-
rounding ECM molecules via receptors such as integrins. Perumpanani contrasts
the terms haptotaxis and chemotaxis, noting that chemotaxis is used to describe
cell motility caused by responses to gradients in soluble attractants, while hapto-
taxis describes motility towards insoluble, substratum-bound attractants such as
laminin and fibronectin.
Perumpanani et al. [107] developed a mathematical model for the invasive pro-
cess in one space dimension. They considered the way in which chemotaxis and
haptotaxis cooperate to regulate invasion. The model is comprised of the partial
differential equations
∂u
∂t=
cell division︷ ︸︸ ︷k1u(k2 − u)− ∂
∂x
chemotaxis︷ ︸︸ ︷[k3ψ(s)u
∂s
∂x+
haptotxis︷ ︸︸ ︷k4χ(c)u
∂c
∂x
],
∂c
∂t= −
proteolysis︷︸︸︷k5pc ,
∂s
∂t=
proteolysis︷ ︸︸ ︷k5k6pc +h(p, s) +
diffusion︷ ︸︸ ︷Dx
∂2s
∂x2,
∂p
∂t=
MMP-2 production︷︸︸︷k7uc − k3pu− k9p︸ ︷︷ ︸
MMP-2 degradation
+
diffusion︷ ︸︸ ︷Dp
∂2p
∂x2,
14
where u(x, t), c(x, t), p(x, t) and s(x, t) represent cells, intact fibronectin, matrix
metaloprotease-2 and the matrix metaloprotease-2 digested soluble fibronectin re-
spectively. The functions ψ(s) and χ(c) are the coefficients of chemotaxis and
haptotaxis. The proteolysis of fibronectin is proportional to the interaction be-
tween protease p and fibronectin c. The term h(p, s) is used to represent the
action of proteases.
Based on numerical simulation of this model, the authors obtained that by
increasing chemotactic sensitivity the invasiveness of cells will decrease. There is
dependence of the invasion speed on MMP-2 production with and without h(p, s).
In all cases, the authors concluded that the solution is a traveling wave moving
in the positive x-direction and the model predicts that, counterintuitively, if the
chemotactic coefficient k3 is increased then the invasiveness will decrease. This
implies that
“ECM chemotaxis actually inhibits invasion, contrary to its conception
as a pro-invasive factor; this is because the gradient in degraded soluble
ECM (s) is in opposite direction to invasion. The predictions of the
model depend crucially on the function of h(p, s).” (p. 2349 in [107]).
If the h(p, s) term is excluded from the model, by increasing the rate of protease
production, k5, then the invasiveness of the cell population will increase continu-
ously. However, if h(p, s) is included, the model estimates a decrease in invasiveness
at high rates of protease production. The model of Perumpanani et al. predicts
that invasion occurs when a cell is between two regions, namely, a haptotactic
gradient of insoluble ECM and a chemotactic gradient of soluble ECM.
In this paper, the competing gradients of digested and undigested fibronectin are
also described. The mathematical model predicts that digested fibronectin retards
cell migration, that is, the cells tend preferentially to the undigested fibronectin
if both the digested and undigested fibronectin are at low concentrations, but as
the concentrations of both these forms of fibronecting are increased, the number
of migrating cells decreases drastically and the migration levels will reduce. The
results of this study show that combining anti-protease therapy with haptotactic
blockade can effectively prevent the unintended augmentation of invasion caused
by antiproteases.
Perumpanani et al. (1999)
Perumpanani et al. in [109] developed and analysed a model for malignant invasion
that combined both proteolysis and haptotaxis. The process of malignant tumour
migration into surrounding tissue can be divided into three main parts: adhesion,
15
proteolysis and migration. The model presented in this paper can be thought of
as a reduced or simplified version of that in [107].
In this paper, the effects on malignant cells of haptotaxis and protease produc-
tion are studied. The authors derived a model for invasion by a combination of
haptotaxis and proteolysis based on a continuum approach in which u(x, t), c(x, t)
and p(x, t) represent the concentration of the invasive cells, ECM and protease,
respectively.
In developing the model of malignant invasion, Perumpanani et al. consider the
movement of the cells spatially, as well as the proliferation of the malignant cells.
They model spatial movement through a description of directed cell migration up
an ECM gradient, represented by ∂c/∂x. This leads to a haptotactic cell movement
term proportional to∂
∂x
(u∂c
∂x
).
The increased proliferation of malignant cells relative to normal cells is assumed
to obey a logistic-type growth. The motility of the ECM is negligible, since the
movement of ECM elements occurs over a much longer timescale than that of
cell migration and protease movement. Hence the authors model the dynamics
of connective tissue by the activity of the tissue protease which is represented by
the term −g(c, p). Perumpanani et al. assume that the protease decays linearly,
with half-life κ, and that protease diffusion is negligible. The authors introduce
the function h(u, c) to represent the dependence of protease production on local
concentration of tumour cells and ECM. Combining the above explanation, they
present the model with the partial differential equations
∂u
∂t=
invasive cell proliferation︷︸︸︷f(u) −
haptotactic cell movement︷ ︸︸ ︷k3
∂
∂x
[u∂c
∂x
],
∂c
∂t=
proteolysis︷ ︸︸ ︷−g(c, p), (1.2)
∂p
∂t=
protease production︷ ︸︸ ︷h(u, c) −
natural decay︷︸︸︷κp ,
where f, g and h are increasing functions of u, c and p.
The authors present a number of simple cases for the functions f, g and h, then
nondimensionalise and simplify the model in the limit of fast protease dynamics
to present a parameter-free, two equation system to model tumour cell invasion.
The model is given by the equations
∂u
∂t= u(1− u)− ∂
∂x
[u∂c
∂x
],
∂c
∂t= −uc2. (1.3)
16
Figure 1.2: The dimensionless tumour cell density resulting from the numerical solution of
the Perumpanani et al. model given by equations (1.3) using MATLAB PDE solver pdepe for
t = 0, . . . , 30, x = 0, . . . , 20, n = 100 spatial points (fine mesh, short domain) and δt = 0.001.
The stability and the numerical solution of this simplified model are analysed,
including a traveling wave analysis. Numerical simulations suggest that the ad-
dition to the model of a small amount of cell diffusion results in no significant
change in either the form or speed of the traveling wave solution corresponding
to invasion, except that the discontinuity in the derivative of u is lost. Numerical
simulations are however easier to calculate with the addition of diffusion.
We present some numerical results for equations (1.3) using MATLAB’s in-built
PDE solver pdepe. Figure 1.2 describes the migration of tumour cells at various
times over the domain x ∈ [0, 20] with 100 spatial discretisation points. It shows
that there are some oscillations in the solution around the left end of the domain.
These oscillations remain throughout the integration and are not dampened as t
increases, however these oscillations will decrease if we use a longer domain spatial
domain (see Figure 1.3).
In Figure 1.3 we demonstrate the numerical result taken from pdepe using a
17
Figure 1.3: The dimensionless tumour cell density resulting from the numerical solution of
the Perumpanani et al. model given by equations (1.3) using MATLAB PDE solver pdepe for
t = 0, . . . , 30, x = 0, . . . , 40, n = 40 spatial points (coarse mesh, long domain) and δt = 0.001.
domain of x ∈ [0, 40] and 40 spatial discretisation points leading to less oscillatory
results than those observed for the shorter domain in Figure 1.2. Figure 1.4 gives
even better results. In this simulation we use the same domain and time length as
in Figure 1.3, however here the mesh is refined to include n = 100 spatial points.
Comparing the depth of migration at t = 30 for the simulations with n = 40 and
n = 100, the front of tumour cells nearly reaches a point x = 40 in the first case,
while for the second, the front has reached only as far as x ≈ 20. This reflects
the restriction that the mesh must be quite fine in order to obtain appropriate
numerical solutions for models of this type using the solver pdepe.
While the solver pdepe can be used to obtain approximate solutions in some
parameter ranges, it is not entirely suited to this hyperbolic PDE model. More
appropriate results for PDE models such as these should be calculated using nu-
merical solvers written by hand.
18
Figure 1.4: The dimensionless tumour cell density resulting from the numerical solution of
the Perumpanani et al. model given by equations (1.3) using MATLAB PDE solver pdepe for
t = 0, . . . , 30, x = 0, . . . , 40, n = 100 spatial points (fine mesh, long domain) and δt = 0.001.
19
Mallet and Pettet (2006)
In this paper, a mathematical model of haptotaxis which includes the adhesion
receptors known as integrins was developed. The model consisted of five species
contributing to the temporal and spatial behaviour of the model system. In the
model, they described both active and inactive integrins as continua in space and
time. The non-dimensionalised model is given by
∂A
∂t= aIE − bA,
∂I
∂t=
∂
∂x
(d1I
∂I
∂x+ dcNI
∂N
∂x− dcηI
∂A
∂x
)− aIE + c(N + dA− I),
∂N
∂t=
∂
∂x
(dNN
∂N
∂x− η̄N ∂A
∂x
)+N(1−N),
∂E
∂t= fN(1−N)− gEP,
∂P
∂t= dP
∂2P
∂x2+ µNE − νP,
where N(x, t), E(x, t), P (x, t), A(x, t) and I(x, t) denoted the density of cellular
material, ECM, protease, and functionally active and functionally inactive inte-
grins respectively.
They found that there exists a traveling wave solution for the model equations
where the minimum wave speed depends on the ratio of integrin binding to integrin
unbinding. The numerical solutions showed a biphasic relationship between the
depth of cell migration and the magnitude of haptotactic response, where the mi-
gration of cells was slowed by small haptotactic coefficients. They also developed a
reduced three-equation model of haptotactic cell migration which gives a good ap-
proximation to the full model. Finally, a travelling wave analysis was investigated
for a further simplified version of the cell migration model.
Eikenberry et al. (2009)
Eikenberry et al. in [49] presented a mathematical model of melanoma inva-
sion incorporating healthy cells and the immune response. They develop a spa-
tially explicit model using partial differential equations to explain the dynamics of
melanoma invasion in the skin. Then, they extend this model to explain the effect
of the immune system and different levels of immune response. To construct this
model, they use the following assumptions.
• Cell motility depends on contact with other cells and oxygen concentration
and it can be modelled using Fick’s Law.
• Oxygen is a growth limiting nutrient and diffuses into the skin from the skin
surface.
20
• The amount of space available and the concentration of oxygen mediate the
proliferation of healthy and cancerous cells.
• Cancer cells that die become necrotic debris.
• Cancer cells produce angiogenic factors.
• Endothelial cells migrate into the system in response to angiogenic factors
and form the tumour vasculature.
• A basement membrane separates the epidermis from the dermis and restricts
cell migration.
The basic model consists of seven variables, namely c, the tumour cell density,
h, healthy cell density, v, the tumour angiogenic factor (TAF) density, b, the
blood vessel endothelial cell density, d, the necrotic debris density, r, the partial
pressure of oxygen, and l the basement membrane (basal lamina) density. Using
the assumptions and variables above, the mathematical model is constructed as a
rather detailed seven equation reaction-diffusion model.
To see the effect of the cellular immune response, they extend the basic model
by introducing two new variables, namely m, the cytotoxic immune cell density
and a, the “immune attracting factor” (IAF) density. In simulations using the
basic model, it can be shown that the tumour spreads throughout the dermis
and forms a significant necrotic core. When an immune response is considered, it
usually inhibits tumour growth, often destroying the invasive tumour or holding the
tumour to a steady state for many years. Immune activation plays the dominant
role in the migration of cells near the primary tumour.
This model provides a segue from the discussion of cell migration models to
a much more closely relevant discussion of mathematical models of the immune
response to growing tumours. This discussion is presented in the following section.
1.3.3 Modelling tumours and the immune system
As has already been noted, it is a fundamental driver for the work of this thesis that
the role of the immune system in responding to growing tumours is still not fully
understood. An understanding of the effects of the immune system on growing
tumours may be aided by constructing and analysing a model of tumour immune
interaction and calculating the parameters of the model using empirical data.
In recent years several researchers have developed mathematical models of var-
ious aspects of the immune response associated with tumour growth [2, 8, 12, 39,
77, 78, 85, 89, 130]. A review of non-spatial mathematical models of tumour and
immune system interactions can be found in Eftimie et al. [48]. Specific aspects
21
include lymphocyte diffusion, proliferation and migration in solid tumours [92].
Tumour growth coupled with immunotherapy has also been investigated using
ODE models [9, 19, 23, 27, 28, 37, 44, 45, 41, 70, 76, 77, 80, 99]. Furthermore,
numerical and bifurcation analyses of the interaction between the growing tumour
and oncolytic virus models have been demonstrated by Novozhilov et al. [103].
Mukhopadhyay and Battacharyya combine this model with the immune system
and analyse results based on a deterministic model [98]. Eikenberry et al. (dis-
cussed above) presented a mathematical model of melanoma invasion into healthy
cells and the subsequent immune response [49]. They developed a spatially explicit
model using partial differential equations to explain the dynamics of melanoma in-
vasion in the skin. They then extended this model to explain the effect of the
immune system in different levels of immune response. Mathematical models of
growing tumours coupled with macrophage response can be found in [17, 97, 104].
de Pillis et al. have developed mathematical models of tumour-immune system
interactions that concentrate on the role of certain effectors in anticancer responses
[37, 39]. A mathematical model describing multi-layered cell growth of a solid tu-
mour under control of an immune response such as that due to tumour-infiltrating
cytotoxic lymphocytes is investigated in [92]. This model also considers the spatio-
temporal dynamics of a solid tumour. de Pillis et al. in [39] focus on the role of
NK and CD8+ T cells in a mathematical model describing the interaction between
a growing tumour and immune system. Furthermore, a model of the interaction
of tumour growth with the immune system using NK cells as the innate immune
system and cytotoxic T lymphocytes as a specific immune system is developed in
[89].
In a study of tumour therapy, Kirkby et al. [75] proposed a mathematical
model of tumour cell response to radiotherapy. Immunotherapy using the cy-
tokine interleukin-2 (IL-2) has had the most impact for cancer treatment [23, 77].
Isaeva and Osipov [67] studied a dynamical model for the tumour-immune response
based on the interactions between interleukin-2 (IL-2) and intercellular cytokine-
mediated responses. They found that, for example, in the case of medium level
antigen presentation, the growth of a tumour depends on the initial tumour size
and the condition of the immune system. Currently, DCs are seen as a potential
mechanism for cancer therapy [25, 81, 118].
Wu et al. proposed a mathematical model that describes the dynamics of in-
teractions of CD8+ T cells and DCs in lymph node, however this model did not
include the growth of a tumour [132]. Most of the work of de Pillis and coauthors
[33, 34, 35, 36, 37, 39] describes a growing tumour interacting with an immune sys-
tem without DCs whereas in [25, 81] DCs are shown to play an important role in
tumour immunotherapy. Castiglione and Picolli in [25] constructed a mathemati-
22
cal model to investigate the effect of tumour immunotherapy, especially dendritic
cell vaccine (DCV), for a generic solid avascular tumour. They applied the theory
of optimal control to find the optimal approach to applying DCV. The model that
they build consists of an immune system including CD4+ T helper and CD8+ T
cells, but this model does not include NK cells.
Below, a more detailed summary is presented for a selection of models of the
interaction between tumours and the immune system. These specific models are
discussed as they form a framework for the ODE-based models to be developed
and analysed in this thesis.
Kirschner and Panetta (1998)
Some research suggests that immunotherapy with the cytokine interleukin-2 may
boost the immune system to fight tumour growth and in this paper due to Kirschner
and Panetta, a mathematical model describing the dynamic interaction between
tumour cells, immune effector cells and IL-2 is illustrated [77]. Also examined are
the effects of adoptive cellular immunotherapy and the circumstances under which
the tumour can be eliminated are described.
To construct the model, Kirschner and Panetta define three populations, namely
the activated immune-system cells (called effector cells), E(t), such as cytotoxic
T-cells, macrophages, and NK cells that are cytotoxic to the tumour cells; the
tumour cells themselves, T (t); and the concentration of IL-2, IL(t). Kirschner and
Panetta describe the interaction between the effector cells, tumour cells, and the
cytokine IL-2 using the model
dE
dt= cT − µ2E +
p1EILg1 + IL
+ s1,
dT
dt= r2(T )T − aET
g2 + T,
dILdt
=p2ET
g3 + T− µ3IL + s2.
The first equation describes the dynamics for the effector cell population. The
Michaelis-Menten form in this equation indicates the saturation effects of the im-
mune response. The average of the lifespan of the effector is 1/µ2. The external
source of effector cells such as LAK and TIL cells is represented by the constant in-
flux term, s1. The second equation explains changes in the tumour cell population
as a result of growth proportional to T and a Michaelis-Menten term reflecting the
limited immune response to the tumour. The final equation gives the dynamics
for the concentration of the IL-2. Again, Michaelis-Menten kinetics are used to
describe the limiting production of IL-2, IL-2 is depleted at a constant rate µ3,
and s2 represents an external input of IL-2 into the system.
23
Kirschner and Panetta analyse the bifurcation behaviour of the model. It is
found that in the absence of s1 the trivial cancer-free steady state, E0, is always
unstable. In the presence of s1, there is a more realistic tumour-free state, (E0, 0, 0).
This implies that effector cells can clear the tumour if the equilibrium is stable. If
the treatment is increased to very high levels then the tumour can be completely
cleared. When s2 is small, the dynamics are found to be similar to those of the
untreated case. If s2 is large, then the effector cells will grow uncontrolled and
the only stable state is (∞, 0, s2/µ3). Finally, they combine both adoptive cellular
immunotherapy and IL-2, finding that the treatment will succeed for tumours of
high antigenicity.
de Pillis et al. (2001)
de Pillis and Radunskaya [34] proposed a competition model of tumour growth
that includes both an immune system and drug therapy, and which uses ordinary
differential equations to describe the system dynamics. The model describing the
interaction between immune cells, tumour cells and host cells and is represented
by the system of differential equations
dI
dt= s+
ρIT
α + T− c1IT − d1I, (1.4)
dT
dt= r1T (1− b1T )− c2IT − c3TN, (1.5)
dN
dt= r2N(1− b2N)− c4TN, (1.6)
where I(t) represents the number of immune cells at time t, T (t) the number of
tumour cells at time t, and N(t) the number of normal, or host, cells at time t.
The source of the immune cells is considered to be outside of the system, so
a constant influx rate, s, is assumed. Next, in the absence of any tumour, the
cells will die off at per capita rate d1. The presence of tumour cells stimulates the
immune response, denoted by the positive nonlinear growth term for the immune
cells ,ρIT
α + T,
where ρ and α are positive constants. Furthermore, the interaction of immune cells
and tumour cells is reflected in both the death of tumour cells and the inactivation
of immune cells, modelled using the terms −c1IT and −c2IT . Proliferation of both
the tumour cells and the normal cells is modelled by a logistic growth law, with
parameters ri and bi denoting the per capita growth rates and reciprocal carrying
capacities of the two types of cells.
The system of equations (1.4)-(1.6) models only the growth of cells and inter-
actions between the immune system, host and tumour. To add the effect of the
24
drug therapy on the system, the function u(t) is used to denote the amount of
drug at the tumour site at time t. The authors assume that the drug response is
represented by an exponential function of the form
F (u) = a(1− e−ku),
where F (u) is the fraction cell kill for a given amount of drug, u, at the tumour
site. Here, a represents the kill rate coefficient for a particular cell type. de Pillis
et al. take k = 1 and then the system with drug interaction is given by
dI
dt= s+
ρIT
ρ+ T− c1IT − d1I − a1(1− e−u)I,
dT
dt= r1T (1− b1T )− c2IT − c3TN − a2(1− e−u)T,
dN
dt= r2N(1− b2N)− c4TN − a3(1− e−u)N,
du
dt= v(t)− d2u.
where the function v(t) represents the rate at which the drug is delivered to the
tumour site and d2 is the decay rate for the drug.
From numerical simulations, it is found that the tumour-free equilibrium without
providing drugs is locally stable if
r1 <c2s
d1
+ c3.
This represents a comparison between the growth rate of the tumour cells, r1, to the
“resistance coefficient”, c2s/d1, which describes how effectively the immune system
competes with the tumour cells. Also, the authors found that the behaviour of the
coexisting equilibrium is very sensitive to the value of ρ, the tumour response rate,
and to s1, the source rate of the immune cells.
In the optimal control model developed as part of this thesis, we will adopt this
kind of modelling strategy and build upon this particular work to incorporate the
different species considered in the preliminary models of this research.
de Pillis et al. (2005)
A mathematical model that describes tumour-immune system interactions, with
a focus on the role of NK and CD8+ T cells is presented in [39]. In their model,
de Pillis et al. note a clear distinction between the dynamics of NK and CD8+
T cells. The authors investigated the parameter estimation and model validation
using data from published mouse and human experimental studies. They also
conducted a sensitivity analysis for the model parameters. To construct this model,
they consider three cell populations, namely tumour population at time t, denoted
25
T (t), total level of NK cell effectiveness at time t, N(t), and L(t) representing the
total level of tumour-specific CD8+ T cell effectiveness at time t. The model is
given by
dT
dt= aT (1− bT )− cNT −D,
dN
dt= σ − fN +
gT 2
h+ T 2N − pNT,
dL
dt= −mL+
jD2
k +D2L− qLT + rNT,
where
D = d(L/T )λ
s+ (L/T )λT.
Here tumour cells proliferate logistically with growth rate a and reciprocal carry-
ing capacity b, are lysed at a rate given by D (explained further below) and are
destroyed by NK cells at a rate c. Natural killer cells are supplied constantly at
a rate σ, die exponentially with rate constant f , and are destroyed in interactions
with tumour cells at rate p and are recruited to the site of the tumour at a rate
proportional to gT 2/(h + T 2). Finally, CD8+ T cells die exponentially with rate
constant m, are destroyed in interactions with tumour cells with rate constant q,
and recruited by NK cells with rate constant r and are also self-recruited at a rate
proportional to jD2/(k +D2).
The somewhat unusual D term is used to represent the effect of enhancing ligand
expression on tumour cells – this is of importance as a framework for constructing
individual cell level interactions for the cellular automata modelling undertaken
as part of the current research. de Pillis et al. found that traditional power-law
representations did not fit well to available experimental results, while on the other
hand, their newly introduced rational law, D, can predict cell lysis quite accurately.
From sensitivity analyses, the authors found that the most significant parameters
in the model were the tumour cell growth rate and λ, the power to which the ratio
of CD8+ T cells to tumour cells was raised in the term D.
Novozhilov et al. (2006)
Novozhilov et al. present a mathematical model of tumour therapy using oncolytic
viruses that specifically target tumour cells – a promising anti-cancer therapeutic
agent [103]. Oncolytic viruses are viruses that specifically infect and kill cancer
cells but not normal cells. Such viruses have been shown to be relatively effective
for reducing or eliminating tumours in clinical trials. To describe the spread of
viral infection of the tumour, the authors introduce a functional response, which is
related to the ratio of uninfected to infected tumour cells. The model of Novozhilov
26
et al. has the form
dX
dt= r1X
(1− X + Y
K
)− bXY
X + Y,
dY
dt= r2Y
(1− X + Y
K
)+
bXY
X + Y− aY,
where X is the size of the uninfected cell population, Y is the size of the infected
cell population, r1 and r2 are the maximum per capita growth rates of uninfected
and infected cells correspondingly, K is the carrying capacity for all cells, b is the
transmission rate of the virus, and a is the rate at which infected cells are killed
by the virus. Rescaling with
x(τ) = X(t)/K, γ(τ) = Y (t)/K, τ = r1t,
leads to the system
dx
dτ= x(1− (x+ y))− βxy
x+ y, (1.7)
dy
dτ= γy(1− (x+ y)) +
βxy
x+ y− δy, (1.8)
where β = b/r1, γ = r2/r1, and δ = a/r1.
The authors conduct a detailed bifurcation and phase plane analysis and it is
found that, in a certain region of the parameter space, both infected and uninfected
tumour cells can theoretically be eliminated in finite time (see Figure 1.5) – that
is, clearance of the tumour as a result of the introduction of the oncolytic virus is
possible.
de Pillis et al. (2009)
de Pillis et al. [33] note that understanding the immune system is essential to
understanding the growth of a tumour and if immunotherapy is to be used in a
clinical setting, its dynamic interaction with chemotherapy and the tumour itself
must be understood. In this paper, the authors update the model of de Pillis et al.
[37] to allow for endogenous IL-2 production, IL-2-stimulated NK cell proliferation
and IL-2-dependent CD8+ T cell self-regulation.
To construct this model, they denote by T (t) the total tumour cell population,
by N(t) the concentration of NK cells per litre of blood, by L(t) the concentration
of CD8+ T cells per litre of blood, by C(t) the concentration of lymphocytes per
litre of blood, not including NK cells and CD8+ T cells, by M(t) the concentration
of chemotherapeutic drug per litre of blood, by I(t) the concentration of IL-2 per
litre of blood, by vL(t) the number of tumour-activated CD8+ T cells injected per
day per litre of blood volume, by vM(t) the amount of doxorubicin injected per
27
Figure 1.5: Phase portrait of the Novozhilov et al. model given by equations (1.7)–(1.8)
using parameter values β = 1.5, γ = 0.7, and δ = 1. A number of sample trajectories in the
uninfected tumour cell–infected tumour cell phase space are presented and we observe for this
parameter set the stability of the trivial steady state, reflecting clearance of the tumour.
28
day per litre of blood volume, and vI(t) the amount of IL-2 injected per day per
litre of blood volume.
The model is given by the relatively comprehensive system of ordinary differen-
tial equations
dT
dt= aT (1− bT )− cNT −DT −KT (1− e−δTM)T, (1.9)
dN
dt= f
(e
fC −N
)− pNT +
pNNI
gN + I−KN(1− e−δNM)N, (1.10)
dL
dt=
θmL
θ + I+ j
T
k + T− qLT + (r1N + r2C)T − uL2CI
κ+ I
−KL(1− e−δLM)L+pILI
gI + I+ νL(t), (1.11)
dC
dt= β
(α
β− C
)−KC(1− e−δCM)C, (1.12)
dM
dt= −γM + νM(t), (1.13)
dI
dt= −µII + φC +
ωLI
ζ + I+ υI(t), (1.14)
where again the authors employ the ratio-form for the CD8+ T cell related killing
of tumour cells, given by
D = d(L/T )λ
s+ (L/T )λ.
Compared with the de Pillis et al. model presented in [37] the authors here
remove the term, gT 2N/(h+T 2), from the model because of its previously observed
insignificance within the context of the model system and due to the additional
complexity it introduces to the form of the model. New terms were added however
such as an IL-2 induced NK cell proliferation term, pNNI/(gN + I).
From the model simulation with no therapy, for large initial tumour size, the
immune system is not able to destroy the tumour and the tumour cell populations
grows to the high tumour equilibrium. However, when the chemotherapy is ini-
tiated, the tumour cell population is rapidly eliminated. For combined therapy,
the tumour cells can also be destroyed, activated CD8+T cells and NK cells drop
slightly but still in agreement with Jurisitc et al. [71]. In this paper, they also
present the sensitivity analysis for model parameters and found that the model is
very sensitive to some of its parameters such as a,KT , δT , u, γ, d, l and s.
The authors suggest that if individual data for CD8+ T cell effectiveness in
killing the tumour can be obtained, it will be possible to determine the feasibility
of using immunotherapy to combat the growing tumour. Their model also indicates
that if the CD8+ T cells kill tumour cells more effectively than immunotherapy,
it may be more useful to give immunotherapy in conjunction with chemotherapy.
29
Whereas, immunotherapy might not be effective in eliminating the tumour cells if
the patient has low immune efficacy.
Wilson and Levy (2012)
In a recent study, Wilson and Levy propose a mathematical model related to
immunotherapy using transforming growth factor β (TGF-β) [129]. They analysed
the effect of anti-TGF-β treatment when combined with a vaccine as treatments
for tumour growth. The model is constructed using the five ordinary differential
equations
dT
dt= a0T (1− c0T )− δ0
ET
1 + c1B− δ0TV,
dB
dt= a1
T 2
c2 + T 2− dB,
dE
dt=
fET
1 + c3TB− rE − δ0RE − δ1E,
dR
dt= rE − δ1R,
dV
dt= g(t)− δ1V,
where T (t) denotes the tumour size, B(t) represents TGF-β concentration, E(t)
denotes activated cytotoxic effector cells, R(t) is regulatory T cells and V (t) is
vaccine-induced cytotoxic effector cells.
The authors numerically solve four cases, namely, no treatment, vaccine treat-
ment, anti-TGF-β treatment and combined treatment. The vaccine treatment and
the anti-TGF-β treatment alone can not eliminate the tumour mass although it
can reduce the tumour size. However, the solution of the model suggests that a
combined anti-TGF-β and vaccine treatment can eliminate the tumour.
From a sensitivity analysis of the model, in the case of combined treatment,
Wilson and Levy found that the system was sensitive to parameters such as the
maximal production rate of TGF-β, a1; the size at which a tumour starts to
produce TGF-β, c2; and the rate of antigenicity, f .
1.3.4 Optimal control applied to tumour-immune system
model
Over approximately the last decade, theoretical researchers have began investi-
gating tumour treatment strategies. In particular, the ordinary differential equa-
tion models discussed already, have been coupled with chemotherapy and/or im-
munotherapy models. Mathematically, this type of investigation falls neatly into
30
the class of problems analysable via optimal control theory. Optimal control the-
ory allows the researcher to investigate a dynamical system (such as a growing
tumour) being affected by some controlling influence (such as a treatment), and
then to optimise the application of that control such that some quantity is op-
timised (such as minimising tumour size at some end-time). In this section, we
review a selection of such models that influence the model development in this
thesis.
de Pillis and Radunskaya (2003)
de Pillis and Radunskaya [36] constructed a mathematical model of tumour growth
and immune system interactions. This model consists of three populations of cells,
namely, I(t) denoting the number of immune cells at time t, T (t) representing
the number of tumour cells at time t, and N(t) denoting the number of normal
or healthy host cells at time t. While this in itself provides a fairly simple three
ODE model of tumour-host interaction, the authors went on to add the effect
of drug therapy to the system where the strategy of application of the drug was
to be solved for. They then applied optimal control theory to solve the problem
for a specified optimal situation. The aim of the optimal control problem is to
determine the chemotherapy administration schedule that minimises the tumour
cell population, while also minimising the destruction of normal cells.
To this end, they attempt to find the control, or drug administration schedule,
v(t), and free final time tf that minimises some objective functional
J(v, tf ) = φ(x(tf , tf ),
subject to the state equations
dI
dt= s+
ρIT
ρ+ T− c1IT − d1I − a1(1− e−u)I,
dT
dt= r1T (1− b1T )− c2IT − c3TN − a2(1− e−u)T,
dN
dt= r2N(1− b2N)− c4TN − a3(1− e−u)N,
du
dt= v(t)− d2u,
and a state constraint
N(t) ≥ 0.75, 0 ≤ t ≤ tf ,
which implies that normal cells must remain above some threshold level required
for a minimal level of patient health. Here u(t) represents the amount of drug
in the system which is administered at rate v(t) and decays naturally with rate
constant d2. The state equations represent the influx, recruitment and produc-
tion of the various cells, the interactions between cells leading to tumour death
31
and neutralisation or removal of healthy and immune cells, and the effects of the
chemotherapeutic drug on each of the cell types.
The authors used a direct collocation method to solve the optimal control prob-
lem over a simulated time period of 150 days. From numerical simulations, they
found that for a weaker immune system, traditional therapy fails to bring the
system to the zero-tumour burden state. However, by applying the drug therapy
using the optimal control solution it is shown that the tumour cell population is
driven to zero, while keeping the normal cell population above the constraint level.
Burden et al. (2004)
Based on the Kirschner and Panetta model [77], Burden et al. [20] developed a
model to determine, using optimal control, under what circumstances a tumour
can be eliminated by an interleukin-based treatment. As described in section 1.3.3,
the model includes the activated immune system cells, denoted x(t), the tumour
cells, denoted y(t), and z(t) to denote the concentration of IL-2. The model has
the form
dx
dt= cy − µ2x+
p1xz
g1 + z+ u(t)s1,
dy
dt= r2(y)y − axy
g2 + y,
dz
dt=
p2xy
g3 + y− µ3z + s2,
where c represents the antigenicity of the tumour, µ2 is the natural death rate of
the effector cells, p1 and g1 affect the shape and rate at which interleukin/effec-
tor cell interactions recruit more immune cells, s1 represents the strength of the
treatment in activating immune cells and u(t) represents the control. The tumour
cells proliferate at a rate given by r2(y) and are killed by immune cells with rate
constant a and at a rate shaped by g2. The interleukin is constantly sourced from
outside the system, decays naturally with rate constant µ3 and is produced by
tumour/immune cell interactions with rate constant p2 and shaping parameter g3.
The objective functional for the problem is defined to be
J(u) =
∫ T
0
[x(t)− y(t) + z(t)− 1
2B(u(t))2] dt,
which reflects the intention to maximise the amount of effector cells and interleukin-
2 as well as to minimise the number of tumour cells and the cost of the control. B
represents a cost weighting factor.
From numerical solutions, run over a 350 day simulation period, it is found that
the cancer cells can be controlled with the tumour reaching a peak between days
200 and 250 (using s1 = 500, B = 5 and c = 0.25). The authors also show the
32
delicate nature of the dynamics with respect to the system parameters, noting that
the cancer cells also can be controlled nearly all of the time, except approximately
at days 320, the tumour starts to grow again because of the unstable dynamics in
the system (using s1 = 500, B = 5 and c = 0.25). Changing only the strength
coefficient for the treatment, namely s1 = 550, the maximum drug level occurs
from days 5 to 100 and is able to keep the cancer within acceptable levels, with
a peak between days 200 to 250. From day 250 until at the end of the simulated
time interval, the tumour cells can be controlled.
Castiglione and Piccoli (2006)
A mathematical model that includes DCs was presented by Castiglione and Pic-
coli [26]. They investigated the effect of tumour immunotherapy using a DCV.
The model consists of five variables representing tumour cells and elements of the
immune system. The model can be written as
dH
dt= a0 − b0H + c0D
[d0H
(1− H
f0
)],
dC
dt= a1 − b1C + c1I(M +D)
[d1C
(1− C
f1
)],
dM
dt=
[d2M
(1− M
f2
)]− e2MC, (1.15)
dD
dt= −e3DC,
dI
dt= a4HD − c4CI − e4I,
whereH denotes the tumour-specific CD4+ T helper cells, C represents the tumour-
specific CD8+ T cells or cytotoxic T lymphocytes, M represents the cancer cells
which expose the tumour-associated antigen, D denotes mature DCs which are
loaded with the antigen and I represents interleukin-2 secreted by CD4+ T cells
and responsible for tumour cell growth. The control variable is added to the fourth
of equations (1.15), and the optimal control problem requires finding u such that
ϕ(x, (T, u)),
with x(0) = x0, is minimised. Here ϕ represents the final cost, T is the terminal
time, x denotes the cell population and x0 is the initial condition. To solve this
problem, they use typical tools of optimal control to find the effect of the DC vac-
cine. They compute the gradient of the cost function with respect to the schedule
and use the steepest descent method for the optimization algorithm. The authors’
numerical results indicate that the treatment is always able to reduce the tumour
burden, however the extent of reduction varies.
33
de Pillis et al. (2007)
de Pillis et al. [40], also investigating tumour-immune system interactions, used
quadratic and linear optimal control problems applied to find strategies for op-
timally administering treatment to a growing tumour. Their model consists of
four populations including drug concentration, similar to their earlier work. The
model describes the growth, death, and interactions of these populations with a
chemotherapy treatment, and is given by
dT
dt= aT (1− bT )− c1NT −KTMT,
dN
dt= α1 − fN + g
T
h+ TN − pNT −KNMN,
dC
dt= α2 − βC −KCMC,
dM
dt= −γM + VM(t),
where T (t) denotes the tumour cell population, N(t) represents the effector-immune
cell population, C(t) is the circulating lymphocyte population and M(t) is the
chemotherapeutic drug concentration.
The authors conduct a stability analysis showing that the T = 0 equilibrium
point is locally asymptotically stable if
a− c1α1γ
γf +KNVM− KTVM
γ< 0.
For other equilibrium points, the equilibrium locations and their stability were
calculated numerically using Maple.
The objective functional employed by de Pillis et al. in this work is given by
J(VM) =
∫ tf
0
(T (t) +
ε
2V 2M(t)
)dt,
and the aim of the optimal control problem, as modelled via this functional, is to
minimise the tumour burden over the time interval while also minimising the total
drug administered. The authors prove existence of an optimal control and also
compare the results using a number of software packages, including Miser3, RIOTS
and DIRCOL. The numerical solutions presented cover a predetermined (rather than
free) time interval of 365 days. From the solution of the optimal control problem,
it is shown that the initial level of drug administration is high for a short period
of time, after which it falls rapidly. In contrast, the two immune cell populations,
that is NK cells and circulating lymphocytes, continue to increase over time. The
tumour population regrows if the low-dose of medicine is shut off. Clinically, this
implies that a patient is required to take low-dose medication regularly for the rest
of his or her life. Both of the models, linear and quadratic control, demonstrated
34
that the best way to fight the tumour is to provide a burst of treatment at the
beginning of the treatment period.
Castiglione and Piccoli (2007)
Another mathematical model describing the interaction between cancer and the
immune system was constructed by Castiglione and Piccoli [27]. They applied
optimal control theory to answer the question of how to decide when and how
much of an anticancer drug to inject so as to induce a prolonged and effective
immune response to the cancer. In this model, the authors use a DCV as an
immunotherapeutic treatment for cancer. The model is similar to that described
in equations (1.15) as developed by the authors in [26]. Castiglione and Piccoli
considered the optimal control problem using three methods, namely continuous-
in-time controls with no jumps, impulsive controls and hybrid controls. From their
numerical solutions, they summarised the results as follows. To reduce the tumour
burden, a high dose of vaccination should be administered early in the treatment
regime, then smaller doses of vaccination should be distributed over the remainder
of the treatment. These findings are seen to be similar to those reported by de
Pillis et al. above.
Cappucio et al. (2007)
Cappuccio et al. [24] also described a mathematical model related to cancer im-
munotherapy. They produced a model that used an optimal control approach to
attempt to minimise the tumour growth and the toxicity of the drug to the patient.
Unlike the previous model due to Castiglione and Piccoli [27], which optimised only
the dosage of the vaccine to be administered to the site of the tumour, Cappucio et
al. also optimise the timing of each drug administration. The mathematical model
of the tumour-immune system to which they couple an optimal control problem, is
based on the Kirschner and Panetta model [77]. They investigated three types of
immunotherapy, namely, CTL therapy, IL-2 therapy and combined therapy. From
the numerical analysis of their model, the authors found that the most successful
method to control the tumour growth, keeping the tumour size below a stated
threshold, is the combined CTL/IL-2 therapy. For monotherapy, IL-2 treatment
is found to be less effective than CTL treatment.
Ghaffari and Naserifar (2010)
In 2010, Ghaffari and Naserifar [61] applied optimal control to a mathematical
model of tumour-immune system interactions, again based on the Kirschner and
Panetta model [77]. In their optimal control problem, Ghaffari and Naserifar
35
minimised the number of tumour cells during and at the end of the treatment.
Such a strategy involves the definition of an objective functional of the form
J(u) = −y(tf ) +
∫ tf
0
[x(t)− y(t) + z(t)− 1
2B(u(t))2
]dt.
where compared with the objective functional used by Burden et al. [20]. For
example, Ghaffari and Naserifar added the term
−y(tf ),
which introduces the possibility to minimise the tumour cell population at the end
time in addition to the burden over time.
Numerical solutions indicate that the addition of this term to the objective
functional allows a treatment strategy to be determined that causes the tumour
cells to be eliminated in a significantly shorter time than in the work of Burden et
al. with identical parameter values. Furthermore, the peak of tumour cells during
therapy is smaller than that seen in [20].
1.3.5 Modelling tumours and the immune system using
cellular automata
One of the earliest hybrid cellular automata models of tumour growth was con-
structed by Ferreira et al [52]. Ferreira et al. produced a mathematical model
for the growth of an avascular tumour, including cell proliferation, motility, and
death, each of which is incorporated via a discrete, individual cell-based cellular
automata. The hybrid nature of the model is due to the inclusion of continuous
reaction-diffusion equations, used to model the concentration of nutrients that are
considered essential and nonessential for cell proliferation. These nutrient concen-
tration equations have the dimensionless forms
∂N
∂t= ∇2N − α2Nσn − λNα2Nσc,
∂M
∂t= ∇2M − α2Mσn − λMα2Mσc,
where N and M represent the concentrations of the two nutrients which diffuse in a
Fickian manner, are consumed by tumour cells and also by healthy host cells. The
equations are coupled with boundary conditions representing a constant source of
nutrients via a blood vessel.
In the discrete cellular automata component, each tumour cell can undergo
division, migration or death. The probability for cancer cell division is given by
Pdiv(x) = 1− exp
[−(
N
σcθdiv
)2],
36
where θdiv controls the shape of the probability curve. The probability of cancer
cell migration takes the form
Pmov(x) = 1− exp
[−σc
(M
θmov
)2],
and the probability for cancer cell death is
Pdiv(x) = exp
[−(
M
σcθdel
)2].
From numerical simulations, it is found that three patterns of tumour growth can
be produced by the model, namely, compact tumours, ramified morphologies and
disconnected patterns.
Based on this model, Mallet and de Pillis [89] presented a mathematical model
of tumour immune-system interactions using a hybrid cellular automata model. In
this model, the authors also employed reaction-diffusion partial differential equa-
tions to describe the distributions of nutrient species which are involved in the
interaction between the immune system and the growing tumour. However they
added to the Ferreira work by incorporating an immune system. In the absence of
the immune system, the authors produced similar results as in the work of Ferreira
et al.. That is, for lower nutrient consumption rates more compact tumours were
simulated, whereas for higher consumption rates ‘branchier’ morphologies were
produced. Also, they found that at the beginning of the growth period the tumour
cells grow exponentially, then linearly until a steady-state of existence is reached.
The numerical simulations demonstrated that there is oscillatory behaviour for
the tumour and immune system cell populations which depend on the strength
of the immune system parameters. Under certain conditions, the tumour can be
eliminated, slightly oscillatory or grow unstable.
1.3.6 Summary
This brief literature review indicates that there is a significant body of research
in the area of tumour growth incorporating interactions with host tissue and in-
deed more recently with the immune system. This review also suggests that the
mathematical modelling of tumour-immune system interactions with specific focus
on cells of the innate (NK cell and DCs) and specific immune system has not yet
been fully studied. The new models developed as part of this research will expand
upon existing models to provide a more complete picture of the interaction between
growing tumours and the host immune system, especially with regard to individual
level cell interactions and the potential for immunotherapeutic treatments.
37
1.4 Thesis structure
The remainder of the thesis comprises three chapters where the primary contribu-
tion to the field of research are presented. In particular, three new mathematical
models of tumour growth and immune system interactions are presented.
Chapter 2 describes a mathematical model of the interaction between a growing
tumour and cells of the innate (NK cells and DCs) and specific (CD8+ T cells)
immune system. To describe this interaction, the model is comprised of four ordi-
nary differential equations (ODEs). We analyse the stability of the model as well
as the simple bifurcation behaviour. Numerical solution of the model is also carried
out and allows for an investigation of the effect of DCs and other components of
the immune system on the tumour cell population. Two mathematical models are
presented here to give a different overview of the model. Before a more complex
mathematical model is provided, first the preliminary model is constructed to serve
as an introductory example to build upon. This model is extended for the optimal
control model that will be presented in Chapter 3.
In Chapter 3, a DCV is added to the mathematical model that is built in the
previous chapter. The aim of this model is to find the optimal DCV to be adminis-
tered to the patient that minimises the tumour burden. To solve this problem, we
apply optimal control theory to a modified version of the model given in Chapter
2. The proof of the existence of an optimal control is presented in this chapter.
We solve the optimal control model using the forward-backward sweep method,
which is also explained in the text. The effect of varying the amount of DCs to be
administered to the tumour site is demonstrated.
Furthermore, a hybrid cellular automata model is presented in Chapter 4 to de-
scribe the interactions between tumour cells and the immune system in much more
detail. This model allows us to investigate the complexity of the biological system,
including cell-cell interactions of every single cell in the system. We include the
effect of chemokines in the model by introducing two partial differential equations
to describe their concentrations and changes due to interactions with the cellular
species in the model. Then, stochastic rules based on descriptions of the various
cell types comprising the host-tumour environment are introduced to the model
via a cellular automata modelling strategy. We use the CA model to simulate the
growth of a tumour in a number of computational “cancer patients” and present
outcomes in a similar manner to Kaplan-Meier survival curves reported in clinical
studies. We define “death” of a patient as the situation where the cells of the tu-
mour reach the boundary of our model domain which effectively represents tumour
metastasis.
Finally, Chapter 5, summarises the results of the thesis presented in Chapter
2-4. Before we conclude with a discussion for future research, contributions of this
38
research are presented in this chapter.
Chapter 2
ODE models of specific immune cell interactions
with a population of tumour cells
2.1 Introduction
In this chapter, a new mathematical model is constructed to provide a description
of the interaction between a growing tumour and cells of the innate and specific
immune system. Commonly, NK cells and CD8+ T cells are included in mathemat-
ical models, because both of these cells can lyse tumour cells [15, 59]. However, in
recent years, it has been reported that DCs can also lyse tumour cells [81]. Some
tumours present DCs and the presence of such cells has a potential role in tumour
control. In this model, we assume that NK and DCs, (the innate immune system),
and CD8+ T cells (the specific immune system) can kill tumour cells. To describe
this interaction, the model is comprised of four ordinary differential equations. We
analyse the stability of the model as well as the simple bifurcation behaviour. The
numerical solution of the model is also considered. Before a more complex math-
ematical model is provided, a preliminary model is first constructed to serve as
an introductory example to build upon. This model also forms the basis for the
optimal control model that will be presented in the following chapter of this thesis.
As has already been noted in the introductory chapter, there is evidence that
the immune system is capable of recognising and eliminating tumour cells [16, 31,
106, 122]. Because of this, much research has been done regarding the activation
of the anti-tumour response and stimulation of the immune system with vaccines
or by direct injection of T cells or cytokines.
Currently, the mechanisms involved in immune system interactions with tumour
cells are still not fully understood. Mathematical models can provide important
tools and information to support the early stages of development of biological tu-
mour therapy. In particular, the model developed here is focused on the attack of
39
40
Figure 2.1: The role of DCs in killing tumour cells (reproduced from [81]).
tumour cells by immune cells and how the behaviour of the tumour cells is changed
because of the presence of these immune cells. We present a simple mathemati-
cal model of a growing tumour cell population incorporating an immune system
comprised of NK cells, DCs and CD8+ T cells.
In recent years, DCs have been identified as an important component in con-
trolling tumour growth. Also DCs have a potential role in directly killing cancer
cells, besides their primary role in the regulation of the adaptive antitumour im-
mune response [81]. This mechanism can be described as in Figure 2.1. Dendritic
cells can stimulate resting NK cells, which in turn, after activation, may induce
dendritic maturation. However, NK cells also can kill immature DCs in peripheral
tissues [96].
This research presented in this chapter proposes a mathematical model to de-
scribe the dynamics of interaction of the growing tumour and the immune system.
We focus on an immune system comprised of NK cells, DCs and CD8+ T cells. Wu
et al. propose a mathematical model that describes the dynamics of interactions of
CD8+ T cells and DCs in the lymph node, however this model does not include the
41
growth of the tumour. Most of the de Pillis papers such as [33, 34, 35, 36, 37, 39]
present mathematical models of a growing tumour interacting with the immune
system without DCs whereas experimental studies (see for example [81]) indi-
cate that DCs play an important role in the modelled tumour immunotherapy.
Castiglione and Picolli in [25] construct a mathematical model to investigate the
effect of tumour immunotherapy, especially the dendritic cell vaccine (DCV), for
a generic solid avascular tumour. They apply the theory of optimal control to
find the optimal DCV. The model that they build consists of an immune system
comprised of CD4+ T helper and CD8+ T cells, but this model does not include
NK cells.
The approach adopted in this thesis is to build a mathematical model describing
the dynamics of interaction between a growing tumour and the immune system.
This model consists of four populations: tumour cells and three components of the
immune system, namely DCs, NK cells and CD8+ T cells. In the subsequent work,
we investigate the effect of DCs and hence, the immune system more generally, on
the growing tumour in more detail than has been described in the literature above.
Beside their function as an antigen presenting cells, DCs can also kill tumour cells
directly. DCs can also be killed by NK cells and CD8+ T cells. Such a mathematical
model of tumour-immune system interactions which specifically considers cells of
the innate (NK cells and DCs) and specific immune systems (CD8+ T cells) has
not yet been fully studied.
This chapter is organised as follows. In Section 2.2, we construct a simple
mathematical model describing the dynamics of interaction between a population
of tumour cells and three components of the host immune system. Then, steady
state and stability analysis of the model is presented and discussed. In Section 2.3,
a more complex model, that includes saturation effects regarding immune system
recruitment, is presented and analysed in a similar manner to the introductory
model. Finally, we discuss the results of both models in Section 2.4.
2.2 Initial mathematical model
The model consists of four populations, namely tumour cells, NK cells, DCs and
CD8+ T cells. The model is described using a series of coupled ordinary differential
equations where the populations of each cell type at time t are denoted by:
• T (t), tumour cells,
• N(t), NK cells,
• D(t), DCs,
42
• L(t), CD8+ T cells.
To construct our initial mathematical model describing a growing tumour inter-
acting with the immune system, we use assumptions based on knowledge of the
immune system and some assumptions stated in [35]. These assumptions can be
summarised as follows.
• The growth of the tumour cell populations is assumed to be logistic in the
absence of an immune response [35].
• NK cells, CD8+ T cells and DCs can kill the tumour cells [43, 72, 81].
• As part of the innate immune response, NK cells and DCs are normally
present in the body, even when no tumour cells are present [37].
• After some number of interactions with tumour cells, NK cells and CD8+ T
cells become inactive [3].
• Dendritic cells can prime and increase the activity of NK cells and CD8+ T
cells [30, 51, 58, 81, 96].
• NK cells can kill immature DCs [96].
• Mature CD8+ T cells can clear out the DCs [25, 132].
Using these assumptions, the model takes the form
dT
dt= aT (1− bT )− (c1N + jD + kL)T, (2.1)
dN
dt= s1 − c2NT + d1DN − eN, (2.2)
dD
dt= s2 − f1LD − d2DN − gD, (2.3)
dL
dt= f2DT − hLT − iL, (2.4)
where a represents the per capita growth rate of tumour cells, b−1 denotes the car-
rying capacity of tumour cells, c1 and c2 are the competitive coefficients between
tumour cells and NK cells, d1 and d2 are the competitive coefficients between NK
cells and DCs, f1 represents the rate DCs are killed by CD8+ T cells, f2 represents
the rate of CD8+ T cell expansion due to the interaction between DCs and tumour
cells, s1 and s2 are the constant sources of NK cells and DCs respectively. Pa-
rameters e, g and i are the per capita death rates of the NK cells, DCs and CD8+
T cells respectively and j is the coefficient of competition between tumour and
DCs. Finally, k denotes the coefficient of competition between tumour and CD8+
T cells.
43
Proliferation of the tumour cells is modelled by a logistic growth law, with
parameters a and b denoting the per capita growth rates and reciprocal carrying
capacities of the tumour cells. The competition of NK cells and tumour cells can
result in either the death of tumour cells or inactivation of the NK cells, resulting
in two competition terms: −c1NT and −c2NT . This form of competition also
occurs between CD8+ T cells and tumour cells resulting in the terms: −kLT and
−hLT . The competition of NK cells and DCs can result in activation of the NK
cells, represented by d1ND, also NK cells can kill DCs, d2ND. The mature CD8+
T cells can clear out the DCs, and the interaction between DCs and tumour cells
can activate the CD8+ T cells, resulting in terms −f1DL and f2DT .
The general initial conditions of equations (2.1)–(2.4) are
T (0) = T0, N(0) = N0, D(0) = D0, L(0) = L0, (2.5)
where each of these initial values is non-negative and specific values are introduced
later when numerical solutions are presented.
We have developed a new ODEs model describing the interaction between a
growing tumour and the immune system. This model is not the same as that
models described in Chapter 1. Kirschner and Panetta model [77] and Wilson and
Levy model [129] proposed a mathematical model describing a growing tumour and
the immune system, whereas these models do not include DCs. Most of de Pillis
papers such as [33, 34, 35, 36, 37, 39] present mathematical models of a growing
tumour interacting with the immune system without DCs. Experimental studies
(see for example [81]) indicate that DCs play an important role in the modelled
tumour immunotherapy. The model constructed by Castiglione and Picolli in [25]
described a mathematical model of specific tumour cells interaction with immune
system include the DCs, however our model describe more detail in DCs popula-
tion. The model that we construct in this section also is not the same as other
models in the literature.
2.2.1 Non-dimensionalisation
To facilitate model analysis, we non-dimensionalise the system as follows. Let the
non-dimensionalised state variables be
T̂ = bT, N̂ =d2
gN, D̂ =
d1
gD, L̂ =
f1
gL, t̂ = gt.
and the corresponding parameters be
a∗ =1
ga, c∗1 =
c1
d2
c1, c∗2 =1
bgc2, e∗ =
1
ge, i∗ =
1
gi,
f ∗2 =f1
bd1gf2, h∗ =
1
bgh, j∗ =
s2
g2j, s∗1 =
d2
gs1, s∗2 =
d1
gs1.
44
Leaving the other parameters unchanged, and dropping the hats and stars for
notational clarity, the non-dimensionalised system is given by
dT
dt= aT (1− T )− (c1N + jD + kL)T, (2.6)
dN
dt= s1 − c2NT +DN − eN, (2.7)
dD
dt= s2 − LD −DN −D, (2.8)
dL
dt= f2DT − hLT − iL. (2.9)
The general initial conditions of equations (2.6)–(2.9) are
T (0) = T0, N(0) = N0, D(0) = D0, L(0) = L0,
where each of these initial values is non-negative.
2.2.2 Steady state and stability analysis
To understand the behaviour of this model, first we analyse the dynamics of the
system. Let E(T ∗, N∗, D∗, L∗) be an equilibrium point of the system described by
the equations (2.6)–(2.9). At an equilibrium point, we have
dT
dt=
dN
dt=
dD
dt=
dL
dt= 0.
As long as s1 6= 0 and s2 6= 0, there is no trivial equilibrium. That is
E(T ∗, N∗, D∗, L∗) 6= (0, 0, 0, 0).
However, it is found that there are two tumour-free equilibrium points, namely,
E1 = (0, N∗, D∗1, 0) and E2 = (0, N∗, D∗2, 0) where
N∗ =s1
e−D∗,
D∗1,2 =(s2 + s1 + e)±
√(s2 + s1 + e)2 − 4 (es2)
2.
These tumour-free equilibrium points have biological meaning or exist if,
1. (e−D∗) > 0, and
2. (s2 + s1 + e) ≥ 2√es2.
This means that if the death rate of NK cells, e, is greater than the value
ecritical = D∗,
45
and if the source term of NK cells, s1, is greater than the value
scritical1 = 2
√es2 − (s2 + e),
then all of the tumour-free equilibrium points have positive values, otherwise they
do not have biological meaning, that is, the values are negative.
The Jacobian matrix for the linearisation of (2.6)–(2.9) about an equilibrium
point, is given by1− 2aT ∗ − c1N
∗ − jD∗ − kL∗ −c1T∗ −jT ∗ −T ∗
−c2N∗ −c2T
∗ +D∗ − e N∗ 0
0 −D∗ −(L∗ +N∗ + 1) −D∗
f2D∗ − hL∗ 0 f2T
∗ −hT ∗ − i
.Evaluating the Jacobian matrix at the general tumour-free equilibrium point re-
duces it to
J1 =
a− c1N
∗ − jD∗ 0 0 0
−c2N∗ D∗ − e N∗ 0
0 −D∗ −(N∗ + 1) −D∗
f2D∗ 0 0 −i
.Henceforth we let submatrix A11 be defined by
A11 =
[D∗ − e N∗
−D∗ −(N∗ + 1)
].
The eigenvalues of J1 are given by
λ1 = a− c1N∗ − jD∗, λ2 = −i,
and the eigenvalues of A11 which are given by
σ(A11) = {λq| det(A11 − λI) = 0, q = 3, 4} ,={λq|λ2 − tr(A11)λ+ det(A11) = 0, q = 3, 4
}.
Using the Routh-Hurwitz conditions, the eigenvalues of A11 have negative real
parts, that is, Re {σ(A11)} < 0, if tr(A11) < 0 and det(A11) > 0. If all the
eigenvalues of the Jacobian matrix J1 are negative then the tumour-free equilibrium
point is stable.
Since, from the existence conditions, the value of D∗−e < 0, the trace of matrix
A11 , that is, (D∗−e)−(N∗+1) is always negative, and the determinant of matrix
A11 is always positive. This means that the eigenvalues of matrix A11 are always
negative. Furthermore, the tumour-free equilibrium points are stable only if
a− c1N∗ − jD∗ < 0.
46
This means that if the rate of tumour growth, a, is greater than
acritical = c1N∗ − jD∗,
then the tumour cell population grows to the high equilibrium points, otherwise it
can be eliminated.
We now analyse the steady-states when there are no source terms, that is both
s1 and s2 are zero. There is one trivial equilibrium point where all the populations
are zero, namely E(T ∗, N∗, D∗, L∗) = (0, 0, 0, 0). The Jacobian matrix here is given
by a 0 0 0
0 −e 0 0
0 0 −1 0
0 0 0 −i
.Then the eigenvalues of the Jacobian matrix are a, −e, −1, and −i. Therefore,
the steady-state (0,0,0,0) is always a locally unstable saddle point.
The equilibrium points representing a tumour coexisting with the immune sys-
tem, Ecoexist = (T ∗, N∗, D∗, L∗), can be found by solving for the solutions of the
nonlinear simultaneous equations (2.6)–(2.9) with left sides set to zero. In this case,
the values of the equilibrium must be solved for numerically. Similarly, the sta-
bility of the coexisting equilibrium points can be found by numerically calculating
the eigenvalues of the Jacobian matrix.
From the above analysis, it can be concluded that
• If the source terms for both NK cells and DCs are nonzero, then the tumour
cells can be eliminated if the tumour growth rate is less than the critical
tumour growth rate, where acritical = c1N∗ − jD∗.
• If there is no constant source term, that is both of the source terms for NK
cells and DCs are zero, then the tumour-free equilibrium is always unstable
and the tumour grows without bound.
2.2.3 Numerical results
The solutions presented in this section are calculated using the ode45 function
in MATLAB to solve equations (2.1)–(2.4) along with initial conditions (2.14).
Parameter values used in this simulation are given in Table 2.1. Maple is used
to calculate the equilibrium points which must be solved for numerically. In the
simulations shown we use initial values T0 = 100, N0 = D0 = L0 = 1 to investigate
the effects on the growing tumour of initial values corresponding with a weak
immune system.
47
Param. Description Value Source
a Tumour growth rate 4.31× 10−1 [37]
b b−1 tumour carrying capacity 2.17× 10−8 [37]
c1 NK cell tumour cell kill rate 3.5× 10−6 [39]
c2 NK cell inactivation rate by tumour cells 1.0× 10−7 [37]
d1 Rate of dendritic cell priming NK cells 1.0× 10−6 Estimate
d2 NK cell dendritic cell kill rate 4.0× 10−6 Estimate
e Death rate of NK cell 4.12× 10−2 [79]
f1 CD8+T cell dendritic cell kill rate 1.0× 10−8 Estimate
f2 Rate of dendritic cell priming CD8+T cell 0.01 Estimate
g Death rate of dendritic cell 2.4× 10−2 [132]
h CD8+T inactivation rate by tumour cells 3.42× 10−10 [37]
i Death rate of CD8+T cells 2.0× 10−2 [37]
j Dendritic cell tumour cell kill rate 1.0× 10−7 Estimate
k NK cell tumour cell kill rate 1.0× 10−7 Estimate
s1 Source of NK cells 1.3× 104 [79]
s2 Source of dendritic cell 4.8× 102 [132]
Table 2.1: Parameter values and their associated sources used in numerical solutions in the
present research.
Most of the parameters here have been taken from [37] which have been fit-
ted to the experimental data of Diefenbach [43]. Most of these parameters also
have been used for numerical simulation of the modelling of immunotherapy and
chemotherapy of tumours in [37].
From the earlier analysis, it is found that with parameter values as reported in
Table 2.1 the tumour-free equilibrium is
E0 = (0, 3.1839× 105, 3.6992× 102, 0),
which is a stable equilibrium point. Using parameter values from Table 2.1, it is
found that there are no equilibrium points with coexisting tumour and immune
cell populations that have biological meaning.
Figure 2.2 shows the evolution of the tumour cells, NK cells, DCs and CD8+
T cells, over time. The tumour cells grow to the maximum value, then can be
eliminated after a couple of days, see Figure 2.2 (top left). This means that in this
case a small immune system is able to clear the tumour cells. NK cells gradually
increase to the maximum value and remain constant to the steady state. Initially,
48
Figure 2.2: The evolution of tumour cells, NK cells, DCs and CD8+ T cells using parameter
values from Table 2.1 with the model given by equations (2.1)–(2.4), showing a tumour cell
population which grows initially to a peak prior to full immune clearance.
DCs quickly increase to a high value then as soon as the population of CD8+ T
cells increase, the DCs drastically decrease because immature DCs are killed by
CD8+ T cells. This decreases the tumour population followed by decreasing CD8+
T cells and as a result DCs tend to a steady-state indicating that CD8+ T cells
can affect the growth of DCs.
Using parameter values from Table 2.1, bifurcation occurs at c1 ≈ 1.3536×10−6.
After bifurcation, tumour cells go to the high tumour equilibrium, even if we start
with one tumour cell. However, before bifurcation the tumour cell population can
be eliminated, as shown in Figure 2.2 (top left).
The evolution of the tumour cells, NK cells, DCs and CD8+ T cells, over time,
with c1 = 3.5 × 10−7 (lower than the critical value for clearance of the tumour
cells) can be seen in Figure 2.3. In this case, the tumour cells cannot be destroyed.
There is oscillation at the beginning of the growth period, then the tumour cell
population remains constant after approximately 500 days.
49
Figure 2.3: The evolution of tumour cells, NK cells, DCs and CD8+ T cells using parameter
values from Table 2.1 except for c1 = 3.5× 10−7, with the model given by equations (2.1)–(2.4),
showing a tumour cell population which grows initially to a peak prior to an oscillatory response
to the immune system and reaching a final, non-zero equilibrium.
50
DCs play an important role in the activation of the immune system since DCs
can activate NK cells and CD8+ T cells. Figure 2.4 and 2.5 show the effect of
varying the source term for DCs on the resulting population of NK cells and CD8+
T cells. Increasing the source term of DCs causes increases to the NK cells and
CD8+ T cells populations. Furthermore, these can affect the growth of tumour
cells since both NK and CD8+ T cells can lyse tumour cells. Figure 2.5 shows the
quite significant role played by DCs in recruiting CD8+ T cells, especially early
in the tumour growth period. Finally, increasing the source term for DCs will
decrease the growth of tumour cells as shown in Figure 2.6. This indicates that an
external source of DCs might be used as treatment if the patient lacks sufficient
DCs in their body. From these simulations, it is shown that DCs alone are not fully
effective for tumour treatment. It is also necessary that the NK cell population is
high enough for an effective anti-tumour response. Finally, we note that while the
peak of the tumour cell population is decreased, the time to clearance is actually
not altered significantly.
From these numerical results we have found that, if the rate, f1, at which CD8+
T cells kill DCs is big enough, the DC population decreases in very short time.
It can reduce the activation of NK cells and CD8+ T cells that cause the tumour
cells to grow to the higher value. This result can explain that CD8+ T cells can
inhibit the function of DCs as an antigen presenting cell (see Figure 2.7).
While we have obtained quite useful results with this simple, initial model, it
does have its shortcomings. An important drawback is that if the rate at which
NK cells kill DCs is very small or the constant source term of DCs is large enough,
the population of NK cells will grow in an uncontrolled manner (see Figure 2.8).
This problem can be avoided by introducing to the rate of NK cell activation, a
saturation effect of DCs (rather than a simple linear interaction). We revise our
model in the next section to overcome these inconsistencies.
2.3 A model incorporating Michaelis-Menten dy-
namics
In this section, we build on the findings of the previous section to construct a more
complex mathematical model that incorporates saturation effects in a number of
dynamic interactions. In particular, the kinetic terms used to model the activation
of NK cells, the recruitment of DCs and the activation of CD8+ T cells will all be
modified from the linear interaction terms of the previous section and instead be
described using Michaelis-Menten saturating kinetics.
As described in section 2.2.3, in certain parameter regimes, the NK cell popu-
lation can grow in an uncontrolled manner. To overcome this problem, here we
51
Figure 2.4: The evolution of NK cells showing the effect of varying the source term of DCs
in numerical solutions of the model given by equations (2.1)–(2.4). Increasing the source term of
DCs increases the peak NK cell population due to the NK cell recruitment role played by DCs
and leads to more effective tumour cell clearance. Other parameters for these simulation taken
from Table 2.1, with values of s2 as indicated on the graph.
52
Figure 2.5: The evolution of CD8+ T cells in numerical solutions of the model given by
equations (2.1)–(2.4), showing that increasing the source term of DCs increases the peak CD8+
T cell population again due to the role played by DC-CD8+ T cell interactions in activating the
T cells. The effect here is more pronounced than it was for NK cells. Other parameters for these
simulation taken from Table 2.1, with values of s2 as indicated on the graph.
53
Figure 2.6: The evolution of tumour cells in numerical solutions of the model given by
equations (2.1)–(2.4), showing the effect of varying the source term of DCs. Increasing the
source term of DCs decreases the peak tumour cell population not only due to an increased DC
population but also the increases in NK and CD8+ T cells observed in Figures 2.4 and 2.5. Other
parameters for these simulation taken from Table 2.1, with values of s2 as indicated on the graph.
54
Figure 2.7: The evolution of tumour cells in numerical solutions of the model given by
equations (2.1)–(2.4), showing the effect of varying the rate at which CD8+ T cell kill DCs, f1,
with s1 = 5.000. Other parameters for these simulation taken from Table 2.1, with values of f1
as indicated on the graph.
55
Figure 2.8: The evolution of NK cells of the model given by equations (2.1)–(2.4) show the
population of NK cell growing in an uncontrolled manner.
56
include saturation of the effect due to DCs in equation (2.11) such that NK cells
are activated by DCs through a term of the form
DN
m1 +D.
By adjusting the value of m1, the rate of NK cell population growth can be con-
trolled. This form is a modification of the familiar Michaelis-Menten term (see for
example [77, 79, 37]).
Similarly, the recruitment of DCs is affected by the number of tumour cells and
this will now be reflected by a Michaelis-Menten term included in equation (2.12),
namelyDT
m2 + T.
CD8+ T cells are also activated by DCs that have come in contact with tumour
cells and this is incorporated in the model through a further Michaelis-Menten
term,DT
m3 + T,
to provide saturation in the effect of the tumour cells (see equation (2.13)). As a
result of these changes, the revised model can be written as
dT
dt= aT (1− bT )− (c1N + jD + kL)T, (2.10)
dN
dt= s1 − c2NT + d11
DN
m1 +D− eN, (2.11)
dD
dt= s2 + l
DT
m2 + T− f1LD − d2DN − gD, (2.12)
dL
dt= f22
DT
m3 + T− hLT − iL, (2.13)
where d11/m1 represents the maximum rate of NK cell activation, l/m2 represents
the maximum dendritic cell recruitment rate by tumour cells, f22/m3 denotes the
maximum rate of CD8+ T cell activation. Parameters m1, m2 and m3 control the
steepness of the NK cell activation curve, the tumour cell recruitment curve and
the CD8+ T cell activation curve, respectively. The remaining parameters are the
same as in Section 2.2.3.
The general initial conditions of equations (2.10)–(2.13) are, as in the previous
section,
T (0) = T0, N(0) = N0, D(0) = D0, L(0) = L0, (2.14)
where each of these initial values is non-negative and where specific values will be
introduced in the numerical solutions.
57
2.3.1 Steady state and stability analysis
To understand the behaviour of this model, first we analyse the dynamics of the
system. Let E(T ∗, N∗, D∗, L∗) be an equilibrium point of the system described by
the equations (2.10)–(2.13). At an equilibrium point, we have
dT
dt=
dN
dt=
dD
dt=
dL
dt= 0.
Provided s1 6= 0 and s2 6= 0, again we see that there is no trivial equilibrium. That
is
E(T ∗, N∗, D∗, L∗) 6= (0, 0, 0, 0).
However, for this revised model, it is found that there is only one tumour-free
equilibrium point, namely, E01 = (0, N∗, D∗, 0) where
D∗ =s2
d2N∗ + 1,
N∗ =(d2s1m1 + d11s2 − em1 − es2) +
√A
2d2em1
,
and
A = (d2s1m1 + d11s2 − em1 + es2)2 + 4 (ed2em1) (s1m1 + s1s2) .
The Jacobian matrix of the linearisation of system (2.10)–(2.13) about an equi-
librium point is given by
B −T ∗ −jT ∗ −T ∗
−c2N∗ −c2T
∗ +d11D
∗
m1 +D∗− e m1d11N
∗
(m1 +D∗)20
m2lD∗
(m2 + T ∗)2−d2D
∗ C −f1D∗
m3f22D∗
(m3 + T ∗)2− hL∗ 0
f22T∗
m3 + T ∗−hT ∗ − i.
,
where
B = (a− 2aT ∗ − c1N∗ − jD∗ − L∗) ,
and
C =
(lT ∗
m2 + T ∗− f1L
∗ − d2N∗ − 1
).
The Jacobian matrix about the tumour-free equilibrium point, E01 = (0, N∗, D∗, 0),
is given by
J2 =
a− c1N∗ − jD∗ 0 0 0
−c2N∗ d11D
∗
m1 +D∗− e m1d11N
∗
(m1 +D∗)20
lD∗
m2
−d2D∗ −d2N
∗ − 1 −f1D∗
f22D∗
m3
0 0 −i
.
58
Henceforth we let A22 define the submatrix
A22 =
d11D∗
m1 +D∗− e m1d11N
∗
(m1 +D∗)2
−d2D∗ −d2N
∗ − 1
.The eigenvalues of J2 are given by
λ1 = a− c1N∗ − jD∗, λ2 = −i,
and the eigenvalues of A22 are given by
σ(A22) = {λq| det(A22 − λI) = 0, q = 3, 4} ,={λq|λ2 − tr(A22)λ+ det(A22) = 0, q = 3, 4
}.
Using the Routh-Hurwitz conditions, the eigenvalues of A22 have negative real
parts, that is, Re {σ(A22)} < 0, if tr(A22) < 0 and det(A)22 > 0. If all the
eigenvalues of the Jacobian matrix J2 are negative then the equilibrium point is
stable.
If we take d11 < e then the we have
d11D∗
m1 +D∗− e < 0.
Furthermore the trace of matrix A22 is negative, that is(d11D
∗
m1 +D∗− e)− (d2N
∗ + 1) < 0,
and also the determinant of matrix A22 is always positive. This means that the
eigenvalues of matrix A22 are always negative. Finally, we have then that the
tumour-free equilibrium points are stable if
1. d11 < e
2. a− c1N∗ − jD∗ < 0.
This means that if the rate of tumour growth, a, is less than
acritical = c1N∗ − jD∗,
and the rate of NK cell activation is less than the death rate of NK cells, then
the tumour cell can be eliminated, otherwise it grows to the nonzero equilibrium
value.
The equilibrium point corresponding with coexistence of a tumour cell popula-
tion and the immune system, Ecoexist = (T ∗, N∗, D∗, L∗), can be found by solving
for the solutions of the nonlinear simultaneous equations (2.10)–(2.13) with left
sides set to zero. In this case, the values of the equilibrium must be solved for
numerically. Similarly, the stability of the coexisting equilibrium points can be
found by numerically calculating the eigenvalues of the Jacobian matrix.
59
Param. Description Value Source
d11 Rate of NK cell activation 0.05 Estimate
f22 Rate of CD8+T cell activation 0.01 Estimate
l DC recruitment rate by tumour cells 0.01 Estimate
m1 NK cell activation rate steepness coeff. 1.0×104 Estimate
m2 Tumour cell recruitment rate steepness coeff. 1.0×104 Estimate
m3 CD8+T cell activation rate steepness coeff. 3.0×103 Estimate
Table 2.2: Parameter values used in numerical solution of equations (2.10)–(2.13) in addition
to those in Table 2.1.
2.3.2 Numerical results
Similarly to Section 2.2.3, the solutions presented in this section are calculated
using the ode45 function in MATLAB to solve equations (2.10)–(2.13). MAPLE is
again used to calculate the equilibrium points which must be solved for numerically.
Parameter values used in this section are given in Table 2.1 and Table 2.2.
From the above analysis, it is found that the tumour-free equilibrium for the
tabulated parameters is
E01 = (0, 3.2934× 105, 3.5784× 102, 0),
and this, as demonstrated in the previous section, is a stable equilibrium point.
Using parameter values from Table 2.1 and 2.2, it is found that there are two
equilibrium points with coexisting tumour and immune cell populations that have
biological meaning, specifically
Eu = (7.0431× 105, 1.2122× 105, 9.6188× 102, 4.7320× 102),
which is unstable, and
Es = (4.4741× 107, 2.9000× 103, 1.8710× 104, 5.2996× 103),
which is a stable equilibrium point.
Figure 2.9 shows the evolution of the tumour cells, NK cells, DCs and CD8+ T
cells, over time, using parameters from Table 2.1 and 2.2. The tumour cells grow
to their maximum value and then are eliminated after approximately 40 days, as
shown in Figure 2.9 (top left). NK cells gradually increase to their maximum value,
however, CD8+ T cells move away from the tumour location (decrease) as soon as
the tumour cells are reduced in number. Dendritic cells increase rapidly in very
60
Figure 2.9: The evolution of tumour cells, NK cells, DCs and CD8+ T cells found by
numerically solving the model given by equations (2.10)–(2.13) using parameter values from
Tables 2.1 and 2.2, showing a tumour cell population which grows initially to a peak prior to full
clearance by the combined immune system.
short time and after reaching a peak, decrease again to a steady-state. When the
tumour cells are eliminated, NK cells remain at a maximum level, however DCs
and T cells approach a lower equilibrium value.
Using parameter values from Table 2.1 and 2.2, bifurcation occurs at
c1 = 1.62509× 10−6.
Below the bifurcation values, the tumour cell population reaches the nonzero tu-
mour equilibrium, even if the system starts with one tumour cell (see Figure 2.10).
However, above the critical c1 value, the population of tumour cells that grows
from the initial single cell can be eliminated, as shown in Figure 2.11.
The evolution of the tumour cells, NK cells, DCs and CD8+ T cells, with c1 =
3.5 × 10−7 (a value below the bifurcation value) and l = 0.1 is shown in Figure
2.12. In this case, the tumour cells cannot be destroyed. There is oscillation
61
Figure 2.10: In the numerical solution of the model given by equations (2.10)–(2.13), using
parameters from Table 2.1 and 2.2, except for c1, bifurcation occurs at c1 = 1.62509 × 10−6.
Shown here is that below the bifurcation value, for example c1 = 1.6250× 10−6, one tumour cell,
T (0) = 1, grows to the nonzero tumour equilibrium.
62
Figure 2.11: In the numerical solution of the model given by equations (2.10)–(2.13), using
parameters from Table 2.1 and 2.2, except for c1, bifurcation occurs at c1 = 1.62509 × 10−6.
In contrast to Figure 2.10, shown here is that above the bifurcation value, for example c1 =
1.6251 × 10−6, an initial population T (0) = 1 grows but is then controlled by the immune
system, returning to a stable zero tumour equilibrium.
63
Figure 2.12: Numerical solution of equation (2.10)–(2.13) using data from Table 2.1 and 2.2,
except for c1 = 3.5×10−7 and l = 0.1. Here the tumour grows to a high peak population initially
before the immune system is primed. The immune system is only able to reduce the tumour
population rather than clear it completely.
in the four populations of cells near the beginning of solution time period and
following initial growth of the tumour population to a high level. This is followed
by a steadying of the cell populations which all remain at a constant equilibrium
level after approximately 400 days. Clearly, this numerical solution represents an
example of the case where the immune system is not able to eliminate tumour
cells.
A number of numerical solutions using parameter values from Tables 2.1 and
2.2 but with different initial values for the tumour cell population are presented in
Figure 2.13. The solutions show that the growth of a tumour can be completely
controlled for some initial values. This means that it is possible that even a weak
immune system is able to clear the tumour cells for certain initial populations.
However, with a larger initial population of tumour cells, the immune system
cannot eradicate the tumour cells. In this case, the tumour cells grow to a nonzero
64
Figure 2.13: The evolution of tumour cells given by solutions of equations (2.10)–(2.13)
using parameters from Table 2.1 and 2.2, with different initial values (indicated on the plot).
This graph indicates that there is a critical initial value of the tumour cell population, where all
numerical solutions of the model with initial value below the critical value grow to the stable
zero tumour equilibrium. However, all solutions with initial value above the critical value grow
to the nonzero equilibrium point
.
equilibrium value and remain steady over time.
DCs play an important role in the activation of the immune system since DCs
can activate NK cells and CD8+ T cells. Figure 2.14 (b) and (c) show the effect of
varying the strength of the DC source on the NK cells and CD8+ T cells. Increasing
the DC source term causes increases to the NK cell and CD8+ T cell populations.
Furthermore, such change can affect the population of tumour cells since both of
NK and CD8+ T cells can lyse tumour cells. Finally, increasing the source term
of DCs will decrease the growth of tumour cells as shown in Figure 2.14 (a). This
suggests that an external source of DCs can be used as treatment if the patient
lacks sufficient DCs in their body. These numerical solutions show however that
DCs alone are not fully effective for tumour treatment. It is also necessary that
65
the NK cell population is high enough for an effective anti-tumour response.
Again from the numerical solutions, if the rate at which CD8+ T cells kill DCs,
f1, is above a critical level, the DC population decreases in very short time. This
reduces the activation of NK cells and CD8+ T cells and causes excessive growth of
tumour cells. This result indicates that CD8+ T cells can also inhibit the function
of DCs as antigen presenting cells. Increasing the parameter, l, the dendritic cell
recruitment rate, will cause decreases to the number of tumour cells.
Figure 2.15 (a) shows that increasing the source of NK cells causes decreases to
the number of tumour cells. However, increasing NK cells can result in decreasing
of CD8+ T cells and DCs as shown in Figure 2.15 (b) and (c). Because of this
effect on the other parts of the immune system, NK cells are considered to be a less
appropriate choice for use as a direct cell treatment. This leaves DCs as a better
choice for immunotherapeutic treatment of tumour cells – a concept investigated
in the following chapter.
2.4 Discussion
In this chapter, two mathematical models of the interaction between a growing
tumour and the immune system have been presented. The models describe how
DCs and NK cells, as the innate immune system, and CD8+ T cells, as the specific
immune system, affect the growth of the tumour cell population. While the models
of de Pillis et al. [34, 35, 36, 37, 39, 33] represent the interaction between a growing
tumour and the immune system without DCs, our model includes more detail
through incorporation of three immune system components. This allows for the
investigation of, for example, the effect on tumour growth of the antigen presenting
cell role of DCs.
The first model developed in this chapter was a relatively simple ordinary dif-
ferential equation-based model describing the dynamics of populations of tumour
cells, DCs, NK cells and CD8+ T cells. The first model used only constant, linear
and linear interaction terms to represent system dynamics. Equilibrium, stability
and numerical analysis of this model, which was similar in structure to many of the
existing mathematical models discussed in the introduction and earlier literature
review chapter, allowed an understanding of the dynamics to be developed and
uncovered some shortcomings in a model of this level of abstraction. In the second
model, saturation of effects (such as activation of immune cells by other cells) were
introduced to deal with some of the downfalls of the initial model.
Using the models developed in this chapter, it was possible to investigate the
growth of the tumour cells and the effect of three components of the immune
system (NK, dendritic and CD8+ T cells) on the growth of the tumour cell popu-
66
(a)
(b)
(c)
Figure 2.14: The evolution of (a) tumour cells, (b) CD8+ T cells and (c) NK cells showing
the effect of varying the DC source term. Other parameters for these simulations are taken from
Table 2.1 and Table 2.2, with s1 = 7.000 and values of s2 as indicated on the graph.
67
(a)
(b)
(c)
Figure 2.15: The evolution of (a) tumour cells, (b) CD8+ T cells and (c) DCs showing the
effect of varying the NKs source term. Other parameters for these simulations are taken from
Table 2.1 and 2.2, with s2 = 7.000 and values of s1 as indicated on the graph.
68
lation. A stability analysis of the equilibria of the mathematical models revealed
that under certain conditions, tumour cells can be completely eliminated by the
immune system. This type of behaviour, for both the initial model and the im-
proved model, is shown Figure 2.2 and Figure 2.9. On the other hand, when the
relevant conditions found via the stability analysis, are not satisfied, the tumour
cell population reaches a nonzero equilibrium, indicating coexistence of the tumour
and immune cell populations, as shown in Figure 2.10.
Both of the models demonstrate the compound effects of elements of the immune
system on the dynamics of tumour growth through their effects on other immune
system components. In particular, the role of DCs as an antigen presenting cell
is shown to play an important role in the effector cell activation, enhancing the
immune system activity (in this case, NK cells and CD8+ T cells).
For small sources of DCs and NK cells, the immune system is not able to elim-
inate the tumour cell population, however for large sources of DCs and NK cells,
the tumour cells can be eliminated. Furthermore, in the second model, increasing
the rate of dendritic cell recruitment, l, will decrease the number of tumour cells.
As described in Mazzolini et al. [93], some patients showed increased NK cell
activity after injection of up to 50× 106 DCs. This is represented qualitatively by
our results indicating that an increased DCs source term can cause increases in the
NK cell population which implies increased NK cell activity. Moreover, our result
can show that this also can increase the magnitude of the CD8+ T cell population.
The models analysed in this chapter provide support for investigating the use
of DCs as an immunotherapeutic treatment. In the following chapter, the model
given here will be extended to include a dendritic cell treatment as a tumour
therapy. This problem will be analysed using optimal control theory, using the
model developed in this chapter, to determine an optimal method for administering
DCs as a treatment with a view to minimising tumour burden.
Chapter 3
An optimal control model of dendritic cell treatment
of a growing tumour
3.1 Introduction
In this section, we study an optimal control treatment strategy for a model of the
interactions between a growing tumour and the host immune system. To obtain
the optimal control model we expand the model developed in Chapter 2, consisting
of four ODEs, by adding the control function, u, which represents a possibly time-
varying application of a dendritic cell treatment as well as an optimal control
statement reflecting an aim to minimise some functional of u. A forward-backward
sweep method, which is based on Runge-Kutta scheme, is then used to solve the
optimal control problem.
The aim of this chapter is to construct an optimal control model of the interac-
tion between a growing tumour and the immune system subjected to an externally
supplied dendritic cell treatment. The model again consists of four variables rep-
resenting the populations of tumour cells and three immune cell populations: NK
cells, DCs and cytotoxic T cells. By applying the theory of optimal control, we
find the optimal control strategy for the application of the DC treatment to be
administered while minimising the tumour burden.
Some research related to the mathematical model of tumour immunotherapy
has been undertaken already, including for example [19, 20, 23, 24, 25, 37, 50, 60,
61, 67, 77]. Kirschner and Panetta [77] illustrated the effect of adoptive cellu-
lar immunotherapy through a mathematical model which consists of tumour cells,
immune-effector cells and interleukin-2. This model describes under what condi-
tions the tumour can be destroyed as a result of the therapy. de Pillis et al. [37]
developed a mathematical model describing the growing tumour with combination
immune, vaccine and chemotherapy treatments. In addition, Cappuccio et al. [23]
69
70
introduced a mathematical model of tumour immune system interactions focusing
on NK and T cell immunity, combined with interleukin-21 as an immunotherapy.
However, there are few models which study the effect of DC vaccines on a growing
tumour, for example [25], but these treat the immune system in a different manner
to the present research.
Furthermore, a number of studies (see for example, [20, 27, 29, 46, 40, 50, 55,
56, 60, 61, 68, 91, 94, 126]) have applied an optimal control strategy to determine
a treatment method to obtain the minimum of some cost (such as monetary costs
and harm to the patient) while minimising the tumour cell population. Where
most of these papers described drug and chemotherapy for the tumour treatment,
only a few of them investigated immune cells as a treatment. For example, based
on the Kirschner-Panetta model [77], Burden [20] and Ghaffari and Naserifar [61]
developed an optimal control strategy to minimise tumour burden. Both of their
models attempt to maximise the amount of effector and interleukin-2 cells and
minimise the number of tumour cells and the cost of the control. A review of
optimal control theory in cancer chemotherapy can be found in Swan [125].
Castiglione and Piccoli constructed a mathematical model to describe the in-
teraction between immune system cells and tumour cells. Then, they applied the
optimal control methods to find the optimal quantity of DCs to be administered
to the tumour site. In this research, we focus on immunotherapy-based treatment,
specifically a dendritic cell-based vaccine, since DCs are safe and have minimal
side effects to the human patients. This research is different from Castiglione and
Piccoli model, since in this model, we include NK cells and also describe in more
detail the role of DCs in tumour control.
This chapter is organised as follows. In Section 3.2 an extension of the ODE
model of Chapter 2 is introduced that incorporates a time-varying dendritic cell-
based treatment strategy and also an objective functional that we seek to minimise
in this modelling study. Further, a discussion of a necessary condition for an opti-
mal strategy is presented to set the background for the remainder of the chapter.
The existence of the optimal control is then proved for the model presented in this
chapter. This allows for a presentation of the optimality system that will be solved
numerically. The numerical scheme itself, a forward-backward sweep method, is
then outlined in Section 3.3 along with a number of important numerical solutions
of the optimal control model. Finally, we discuss the results of the model in the
context of the tumour growth problem.
71
3.2 Optimal control problem
Recall from Chapter 2 the revised ODE model (incorporating Michaelis-Menten
dynamics) given by equations (2.10)–(2.13) describing tumour growth and inter-
action with the NK, dendritic and CD8+ T cells of the host immune system. In
particular, with T (t) denoting the tumour cell population, N(t) denoting the NK
cell population, D(t) denoting the dendritic cell population and L(t) representing
the population of CD8+ T cells, the tumour system was described by the equations
dT
dt= aT (1− bT )− (c1N + jD + kL)T, (3.1)
dN
dt= s1 − c2NT + d11
DN
m1 +D− eN, (3.2)
dD
dt= s2 + l
DT
m2 + T− f1LD − d2DN − gD, (3.3)
dL
dt= f22
DT
m3 + T− hLT − iL, (3.4)
along with appropriate initial conditions, governing the time evolution of the four
species of interest. For a description of the individual parameters and terms in the
equations, see Chapter 2.
In this section, we extend this ODE model further to incorporate a representa-
tion of a dendritic cell immunotherapeutic strategy for treatment of the growing
tumour. We consider the problem of controlling the immunotherapy in the most
advantageous way by casting the model as an optimal control problem.
In view of the equations above, we add a dendritic cell treatment to aid in
decreasing the tumour burden by altering equation (3.3). Restating the full model
for later reference, we have
dT
dt= aT (1− bT )− (c1N + jD + kL)T, (3.5)
dN
dt= s1 − c2NT + d11
DN
m1 +D− eN, (3.6)
dD
dt= s2 + l
DT
m2 + T− f1LD − d2DN − gD + wu (t) , (3.7)
dL
dt= f22
DT
m3 + T− hLT − iL. (3.8)
where w is the strength of the treatment and u(t) represents the time varying rate
of application of the dendritic cell treatment, satisfying a ≤ u(t) ≤ b, with a and
b some minimum and maximum rates of cell treatment.
Through the application of the dendritic cell treatment, we seek to minimise the
tumour burden over time while simultaneously minimising the amount of treatment
required. As such, we consider the treatment rate to be a control and cast the model
72
along with these aims as an optimal control problem. The objective functional, to
be minimised here, is then given by
J(u) =
∫ tf
0
(T (t) +
B
2u2(t)
)dt, (3.9)
where tf is the specified final time. The first term represents the tumour cell
burden, to be minimised during the treatment, and the second term reflects a
cost of the dendritic cell treatment (such as monetary costs, harm to patient, etc)
and is also to be minimised. B is a weighting factor that represents the effect
of treatment for patient. The aim is to minimise the number of tumour cells as
well as the cost of vaccine to be administered over all acceptable control functions
u(t). Note that this is not the only acceptable functional, for example the final
tumour cell population could be included to give Jalt(u) = T (tf ) + J(u), however
the objective functional as given in equation (3.9) is consistent with others used
in the literature. This objective functional is the same as objective functional
employed by de Pillis et al. [40].
Hence, we attempt to find an optimal control function u(t) = u∗, in other words
the optimal treatment, such that
J(u∗) = minu∈U{J(u)} ,
where U = {u(t)|a ≤ u(t) ≤ b, t ∈ [0, tf ]} is the control set.
With the optimal control problem defined, we now move to proving the existence
of an optimal control that minimises the given objective functional.
3.2.1 Derivation of necessary condition
In this section, we derive a necessary condition for the optimal control of a system
of the form investigated here. More detail can be found in [82], which forms the
basis for the derivation presented here.
Consider an optimal control problem comprised of ordinary differential equations
x′(t) = g(t, x(t), u(t)),
where x(t) represents the state variable and u(t) is the control variable. The basic
optimal control problem consists of finding a piecewise continuous control function
u(t) and the associated state variable x(t) to maximise (minimise, with reversed
directions on inequalities) the given objective functional. That is, finding x(t) and
u(t) to satisfy
maxu
∫ tf
t0
f(t, x(t), u(t)) dt,
73
subject to
x′(t) = g(t, x(t), u(t)),
where x(t0) = x0 is given and x(tf ) is free.
Suppose u∗ is an optimal control and x∗ the corresponding state. So we have
J(u) ≤ J(u∗) < ∞ for all controls u. Now, let h(t) be a piecewise continuous
variation function and ε ∈ R. Then uε = u∗(t) + εh(t) will be another valid
piecewise continuous control function. Let xε(t) be the state corresponding to
u∗(t) + εh(t). Then we have
d(xε(t))
dt= g(t, xε(t), uε(t)),
and the objective functional at uε is
J(uε) =
∫ tf
t0
f(t, xε(t), uε(t)) dt. (3.10)
Now, let λ(t) be a piecewise differentiable function to be determined. By the
Fundamental Theorem of Calculus we have∫ tf
t0
d
dt[λ(t)xε(t)] dt = λ(tf )x
ε(tf )− λ(t0)xε(t0),
which implies ∫ tf
t0
d
dt[λ(t)xε(t)] dt+ λ(t0)x0 − λ(tf )x
ε(tf ) = 0.
Adding this equation to equation (3.10) gives
J(uε) =
∫ tf
t0
(f(t, xε(t), uε(t)) +
d
dt(λ(t)xε(t))
)dt+ λ(t0)x0 − λ(tf )x
ε(tf )
=
∫ t1
t0
(f(t, xε(t), uε(t)) + λ′(t)xε(t) + λ(t)g(t, xε(t), uε(t))) dt+
λ(t0)x0 − λ(tf )xε(tf ), (3.11)
Since, the maximum of J with respect to u occurs at u∗, then the derivative of
J(uε) with respect to ε is zero, namely,
0 =d
dεJ(uε)
∣∣∣∣ε=0
= limε→0
J(uε)− J(u∗)
ε.
Differentiating equation (3.11) and using the above result gives
0 =d
dεJ(uε)
∣∣∣∣ε=0
=
∫ tf
t0
∂
∂ε
(f(t, xε(t), uε(t)) + λ
′(t)xε(t) + λ(t)g(t, xε(t), uε(t)) dt
)∣∣∣∣ε=0
− ∂
∂ελ(t1)xε(t1)
∣∣∣∣ε=0
.
74
Applying the chain rule to f and g, it follows that
0 =
∫ tf
t0
(fx∂xε
∂ε+ fu
∂uε
∂ε+ λ′(t)
∂xε
∂ε+ λ(t)
(gx∂xε
∂ε+ gu
∂uε
∂ε
))∣∣∣∣ε=0
dt
− λ(tf )∂xε
∂ε(tf )|ε=0.
=
∫ tf
t0
(fx + λ(t)gx + λ′(t))∂xε
∂ε(t)
∣∣∣∣ε=0
+ (fu + λ(t)gu)h(t) dt
−λ(tf )∂xε
∂ε(tf )
∣∣∣∣ε=0
. (3.12)
Now, we want to choose the adjoint function, λ(t), to simplify equation (3.12) by
forcing to zero the coefficients of
∂xε
∂ε(tf )
∣∣∣∣ε=0
,
both in the integrand and in the boundary condition term. Thus, we choose the
function λ(t) to satisfy
λ′(t) = −fx(t, x∗(t), u∗(t))− λ(t)gx(t, x∗(t), u∗(t)), (3.13)
as well as the boundary condition
λ(tf ) = 0. (3.14)
Here, equation (3.13) is referred to as the adjoint equation and equation (3.14)
is called the transversality condition.
Now if we take
h(t) = fu(t, x∗(t), u∗(t)) + λ(t)gu(t, x
∗(t), u∗(t)), (3.15)
then substitute equation (3.13), equation (3.14) and equation (3.15) into equa-
tion (3.12), we have
0 =
∫ tf
t0
(fu(t, x∗(t), u∗(t)) + λ(t)gu(t, x
∗(t), u∗(t)))2 dt,
which implies the optimality condition
fu(t, x∗(t), u∗(t)) + λ(t)gu(t, x
∗(t), u∗(t)) = 0, (3.16)
for all t0 ≤ t ≤ tf . Equations (3.13), (3.14) and (3.16) form a set of necessary
conditions that an optimal control and state must satisfy. In practice, the above
equations need not be re-derived for each application. Furthermore, the above
necessary conditions can be generated via the Hamiltonian of the problem, denoted
H, which is defined as follows
H(t, x, u, λ) = f(t, x, u) + λg(t, x, u).
75
That is, the Hamiltonian is equal to the integrand of the objective functional added
to the product of the adjoint function and the right hand side of the differential
equation system.
So in terms of the Hamiltonian the above necessary conditions can be written
as
Optimality condition:∂H
∂u= 0 at u∗ ⇒ fu + λgu = 0,
Adjoint equation: λ′ = −∂H∂x⇒ λ′ = −(fx + λgx),
Transversality condition: λ(tf ) = 0,
and have the state equation
x′ = g(t, x, u) =∂H
∂λ, x(t0) = x0.
3.2.2 Existence of an optimal control
To prove the existence of an optimal control for the system of equations presented
in Section 3.2, we use the fact that supersolutions T ,N,D,L of
dT
dt= aT ,
dN
dt= s1 + d11N,
dD
dt= s2 + lD + u,
dL
dt= f22D,
are bounded on a finite time interval.
The existence of the optimal control can be obtained by using the following
result due to Fleming and Rishel [54]. Before we present the existence theorem of
the optimal control, the existence theorem of initial value problem is introduced.
The proof of the theorem can be found in, for example, [87].
Theorem 3.1. Given the initial value problem
dx(t)
dt= f(t, x(t)), (3.17)
x|t=τ = ξ.
This problem has a solution if for some
Ra,b = (t, x) : |t− τ | ≤ a, |x− ξ| ≤ b,
76
centred about (τ, ξ) the restriction of f to Ra,b is continuous in x for fixed t, mea-
surable in t for fixed x, and satisfies
|f(t, x)| ≤ m(t), (t, x) ∈ Ra,b,
for some m integrable over the interval [τ − a, τ + a].
Generally speaking, this theorem tells us that the solution to the initial value
problem exists if the function f in equation (3.17) is bounded.
Theorem 3.2. Consider the control problem with system equations (3.5)–(3.8).
There exists an optimal control, u∗ ∈ U such that
minu∈U
J (u) = J (u∗) ,
if the following conditions are satisfied.
(i) The set of controls and corresponding state variables is not empty.
(ii) The control set U is convex and closed.
(iii) Each right hand side of the state system (3.5)–(3.8) is bounded above by a
linear function in the state and control variable.
(iv) The integrand of the objective functional is convex on U and is bounded below
by −c2 + c1u2 with c1, c2 > 0.
Proof. For the given model, we prove each condition in turn.
(i) Using Theorem 3.1 we can prove the existence of solutions of (3.5)–(3.8) with
bounded coefficients.
(ii) The control set is convex and closed by definition.
(iii) The right hand side of the state system is linear in the state and control
variable. We let δ(t,X) be the right hand side of system (3.5)–(3.8) without
the control function u(t). That is,
f(t,X, u) = δ(t,X) +
0
s1
s2 + u
0
, with X =
T
N
D
L
.
77
Using the boundedness of the solutions, we see that
|f(t,X, u)| ≤
∣∣∣∣∣∣∣∣∣
a 0 0 0
0 d11 0 0
0 0 l 0
0 0 f22 0
T
N
D
L
∣∣∣∣∣∣∣∣∣+
∣∣∣∣∣∣∣∣∣
0
s1
s2 + ωu
0
∣∣∣∣∣∣∣∣∣
≤
∣∣∣∣∣∣∣∣∣
0
s1
s2
0
∣∣∣∣∣∣∣∣∣+
∣∣∣∣∣∣∣∣∣
a 0 0 0
0 d11 0 0
0 0 l 0
0 0 f22 0
T
N
D
L
∣∣∣∣∣∣∣∣∣+
∣∣∣∣∣∣∣∣∣
0
0
ωu
0
∣∣∣∣∣∣∣∣∣
≤ C1 (1 + |X|+ |u|) ,
where C1 depends on the coefficients of the system. Hence, the right hand
side of the state equation (3.5)–(3.8) is bounded above by a sum of the state
and control variable.
(iv) To prove that the integrand of the objective functional is convex, we recall
that a real valued function f on an interval is said to be convex if, for any
two points x and y in the interval and for any p ∈ [0, 1],
f(px+ (1− p)y) ≤ pf(x) + (1− p)f(y).
Hence, to prove the integrand of the objective functional is convex, we need
to show that
J(t, T, pu+ (1− p)v) ≤ pJ(t, T, u) + (1− p)J(t, T, v),
where u, v ∈ U , 0 < p < 1. We see the difference of the left and right sides
of the inequality can be written as
J(t, T, pu+ (1− p)v)− (pJ(t, T, u) + (1− p)J(t, T, v))
= T (t) +B
2(v2 − 2pv2 + p2v2 + p2u2 − 2p2vu+ 2pvu)
−(T (t) +
B
2pu2 +
B
2v2 − B
2v2p
)=B
2(p2 − p)(v − u)2.
Since, p ∈ (0, 1), we have (p2−p) < 0 and obviously (v−u)2 > 0, so it follows
that the expression B(p2 − p)(v − u)2/2 is negative. Hence, the condition
J(t, T, pu+ (1− p)v) ≤ pJ(t, T, u) + (1− p)J(t, T, v),
is satisfied. Furthermore, from the objective functional we have
T +B
2u2 ≥ B
2u2 ≥ −c2 +
B
2u2,
which gives −c2 + c1u2 as the lower bound, where c1, c2 > 0.
78
The conditions are satisfied and hence, we conclude that there exists an optimal
control.
The Theorem 3.2 and similar proof can be found in, for example, [20, 40].
3.2.3 Optimality system
Having proved the existence of an optimal control for minimising the functional
(3.9) subject to the equations (3.5)–(3.8), we now derive the necessary conditions
for this particular optimal control using Pontryagin’s maximum principle [111].
Using the functional given in equation (3.9), the Hamiltonian of the system is
given by
H =
[T +
B
2u2
]+
4∑i=1
λigi, (3.18)
where the functions gi are the right hand sides of the state equations (3.5)–(3.8).
Then the Lagrangian is defined as
Lg = H − ω1 (t) (u− a)− ω2 (t) (b− u) ,
where H is the Hamiltonian as defined in (3.18) and ω1 (t), ω2 (t) ≥ 0 are penalty
multipliers satisfying
ω1 (t) (u∗ − a) = 0,
ω2 (t) (b− u∗) = 0.
Hence,
Lg =
[T +
B
2u2
]+ λ1 (aT (1− bT )− c1NT − jDT − kLT )
+ λ2
(s1 − c2NT + d11
DN
m1 +D− eN
)+ λ3
(s2 + l
DT
m2 + T− f1LD − d2DN − gD + wu (t)
)+ λ4
(f22
DT
m3 + T− hLT − iL
)− ω1 (t) (u− a)− ω2 (t) (b− u) ,
Note that at u∗, H = Lg.
Theorem 3.3. Given optimal control u∗ and solutions T ∗, N∗, D∗ and L∗ of the
corresponding state system (3.5)–(3.8), there exist adjoint functions λ1, λ2, λ3 and
79
λ4 satisfying
λ′
1 = −1 + λ1 (2abT − a+ c1N + jD + kL) + λ2c2N − λ3l
(m2D
(m2 + T )2
)+ λ4
(−f22
m3D
(m3 + T )2 + hL
),
λ′
2 = λ1c1T + λ2
(c2T − d11
D
m1 +D+ e
)+ λ3d2D,
λ′
3 = λ1jT − λ2d11m1N
(m1 +D)2+ λ3
(f1L+ +d2N + g − l T
m2 + T
)− λ4f22
T
m3 + T,
λ′
4 = λ1kT + λ3f1D + λ4 (hT + i) .
and λ1(tf ) = λ2(tf ) = λ3(tf ) = λ4(tf ) = 0 (the transversality conditions). Fur-
thermore
u∗ (t) = min
{max
{a,−λ3w
B
}, b
}.
Proof. The form of the adjoint equations and transversality conditions are stan-
dard results from Pontryagin’s Maximum Principle [111]. We differentiate the
Lagrangian with respect to the states, T,N,D and L respectively, and then the
adjoint system can be written as
λ′
1 = −∂Lg∂T
= −1 + λ1 (2abT − a+ c1N + jD + kL) + λ2c2N − λ3l
(m2D
(m2 + T )2
)+ λ4
(−f22
m3D
(m3 + T )2 + hL
),
λ′
2 = −∂Lg∂N
= λ1c1T + λ2
(c2T − d11
D
m1 +D+ e
)+ λ3d2D,
λ′
3 = −∂Lg∂D
= λ1jT − λ2d11m1N
(m1 +D)2+ λ3
(f1L+ +d2N + g − l T
m2 + T
)− λ4f22
T
m3 + T,
λ′
4 = −∂Lg∂L
= λ1kT + λ3f1D + λ4 (hT + i) .
The optimality equation is∂Lg∂u
= 0 at u∗.
Thus,
∂Lg∂u
∣∣∣∣u∗
= Bu+ λ3w − ω1 + ω2 = 0
⇒ Bu∗ = −λ3w + ω1 − ω2.
Hence we have
u∗ (t) =−λ3w − ω1 + ω2
B. (3.19)
80
Now, we consider all possible values of optimal control u∗, of which there are
three cases.
1. For {t : a < u(t) < b}.
In this case, since ω1 = ω2 = 0, then from equation (3.19), the optimal
control is
u∗ (t) =−λ3w
B.
2. For {t : u(t) = b}.
Then, from equation (3.19), we have
b = u∗ (t) =−λ3w + ω1 − ω2
B.
In this case, ω1 = 0, consequently
b = u∗ (t) =−λ3w − ω2
B.
or
b+ω2
B=−λ3w
B.
Since ω2 ≥ 0, then
b+ω2
B≥ b,
Thus,
b = u∗ ≤ −λ3w
B.
So, in this case, we have
u∗ (t) = min
{−λ3w
B, b
}.
3. For {t : u(t) = a}.Then, from equation (3.19), we have
a = u∗ (t) =−λ3w + ω1 − ω2
B.
In this case, ω2 = 0, consequently
a = u∗ (t) =−λ3w + ω1
B,
or
a− ω1
B=−λ3w
B.
Since ω1 ≥ 0, then
a− ω1
B≤ a.
81
Thus,
a = u∗ ≥ −λ3w
B.
So, in this case, we have
u∗ (t) = max
{a,−λ3w
B
}.
Combining these three cases, we conclude
u∗ (t) =
−λ3w/B, a < −λ3w/B < b
a, −λ3w/B ≤ a
b, −λ3w/B ≥ b.
In compact notation,
u∗ (t) = min
{max
{a,−λ3w
B
}, b
}.
3.3 Numerical solutions
In this section we provide a description of the numerical scheme employed to solve
the optimal control problem developed in Section 3.2. We then present a number
of numerical solutions of the model to demonstrate its implementation and possible
outcomes.
3.3.1 Forward-backward sweep method
In this thesis, the numerical solution of the optimal control problem given by
equations (3.5)–(3.8) along with the objective functional equation (3.9), is obtained
using the iterative scheme referred to as the forward-backward sweep method. In
short, the method involves first solving the state system and its associated initial
condition, forward in time. Then, the adjoint system with its initial condition
obtained from the current iteration of the state system is solved backward in time to
meet the transversality condition. An initial guess for the control, u, is updated by
using a convex combination from the optimality equation. The iteration is repeated
until the difference between values of state, adjoint and control variable of the
current iteration and previous iteration are within some acceptable tolerance level.
In this numerical solutions, the ODE is solved using Runge-Kutta method. The
ode45 function in MATLAB can not used any more since the forward-backward
sweep method requires a uniform step size. The parameters used in the numerical
82
solutions in this chapter are taken from Table 2.1 and Table 2.2 in Chapter 2, and
we now expand upon this informal description of the scheme.
To solve our optimal control problem
min
∫ tf
t0
f(t,x(t), u(t)) dt,
subject to
x′(t) = g(t,x(t), u(t)),
with the initial condition x(t0) = x0, we introduce the Hamiltonian function
H(t,x, u, λ) = f(t,x,u) + λTg(t,x, u).
The problem then becomes
x′(t) = g(t,x(t), u(t)), x(t0) = x0,
λ′ = −∂H
∂x= −(fx + λgx), λ(tf ) = 0,
0 =∂H
∂u= fu + λgu, at u∗.
In this research, we use the forward-backward sweep method (as indicated in
[82]) to solve the above problem. The method involves the following iterative
scheme.
1. Discretise the time domain into N+1 equidistant mesh points t1, t2, . . . , tN+1,
where ti = (i − 1)∆t, i = 1, 2, . . . , N + 1 and ∆t = (tf − t0)/(N + 1).
The approximation to the solution is then given by xi = x(ti) and we have
ui = u(ti) and λi = λ(ti), all for i = 1, 2, . . . , N + 1.
2. Make an initial guess for ui over the entire interval.
3. Using the initial condition x1 = x(t0) and the above-mentioned values for ui,
solve x′(t) = g(t,x(t), u(t)) forward in time using the order 4 Runge-Kutta
method. That is, given the initial value problem
x′(t) = g(t,x(t)),
and a step size h, the approximation of x(t+ h) using order 4 Runge-Kutta
method takes the form
x(t+ h) ≈ x(t) +h
6(k1 + 2k2 + 2k3 + k4),
83
where
k1 = f(t, x(t))
k2 = f(t+h
2, x(t) +
h
2k1) (3.20)
k3 = f(t+h
2, x(t) +
h
2k2) (3.21)
k4 = f(t+ h, x(t) + hk3)
4. Using the transversality condition λN+1 = λ(tf ) = 0 and the values for ui
and x, solve λ′ = −(fx + λgx) backward in time.
5. Update ui by entering the new xi and λi values into the optimal control. In
equations (3.20) and (3.21), we calculate k2 and k3 at time t+ h2. Since, u is
a function of time, then we approximate ui+h/2 with the average
ui + ui+1
2.
6. Check convergence where we aim for a relative error to be negligibly small.
That is, ∥∥ui − uoldi
∥∥‖ui‖
≤ η,
where η is the acceptable tolerance, ui represents the current approximation,
uoldi denotes the previous approximation and ‖·‖ refers to the sum of the
absolute value of the terms. To avoid the problem of zero controls, multiply
both sides by ‖ui‖, to give
η ‖ui‖ −∥∥ui − uold
i
∥∥ ≥ 0,
or
η
N+1∑i=1
|ui| −N+1∑i=1
∣∣ui − uoldi
∣∣ ≥ 0.
Similarly, we apply this convergence criteria to the state and adjoint vari-
ables. So for the state variables, we have
ηN+1∑i=1
|xi| −N+1∑i=1
∣∣xi − xoldi
∣∣ ≥ 0,
and for the adjoint variables, we have
ηN+1∑i=1
|λi| −N+1∑i=1
∣∣λi − λoldi
∣∣ ≥ 0.
If the values of the state, adjoint and optimal control variables in this itera-
tion and the last iteration are negligibly close, output the current values as
solutions. Otherwise return to Step 3, using the new ui values.
84
Figure 3.1: The evolution over time of the tumour cell population calculated via the numer-
ical solution of system (3.5)–(3.8) using different strengths for the DCV (legend shown on the
graph), administered to reduce tumour burden. This plot shows that increasing the strength of
the vaccine decreases the magnitude of the peak tumour cell population and also the time taken
to reach the peak.
3.3.2 Simulation and results
Figure 3.1 illustrates the evolution over time of the tumour cell population calcu-
lated via the numerical solution of system (3.5)–(3.8) using different strengths for
the DCV. The parameter values presented in Tables 2.1 and 2.2 in Chapter 2 and
the initial value T (0) = 100, N(0) = 1, D(0) = 1 and L(0) = 1 are again used in
this simulation. With the initial value T0 = 100 cells, without vaccine, tumour cells
grow to a maximum value of approximately 1000 cells within just 12 days. The
remaining numerical solutions show, not surprisingly, that increasing the strength
of the DCV to be administered to the tumour site decreases the number of tu-
mour cells and also shortens the time at which the tumour cell population reaches
its peak level. By adding a significantly strong vaccine, for example 5x106, the
85
Figure 3.2: The evolution of two tumour cell populations using a large initial tumour cell
population, showing the impact of the vaccine-based control strategy. The dashed line shows the
tumour population growth resulting from no control, while the solid line (more easily seen inset)
shows the tumour cell population eradication resulting from the optimal control vaccine strategy.
tumour cells can be cleared in just in 6 days. However, while this demonstrates
the effectiveness of the model, we note that this is somewhat unrealistic as the
vaccine is administered immediately, prior to the tumour cell population reaching
a detectable size.
Figure 3.2 presents a plot of the tumour cell population using a large initial value
of tumour cells, demonstrating the impact of the vaccine-based control strategy.
The dashed line shows the tumour population growth resulting from no control,
while the solid line (more easily seen inset) shows the tumour cell population
eradication resulting from the optimal control vaccine strategy. With initial value
T0 = 42310 cells, without vaccine, tumour cells grow to the nonzero tumour equi-
librium after around 120 days. However this can be eliminated in just under 40
days using the optimal control strategy. The inset plot shows that the evolution of
tumour cells in 40 days where without control, the number of tumour cells remains
86
Figure 3.3: The optimal control, u, used for treatment of the tumour cell population shown
in Figure 3.2.
87
Figure 3.4: The evolution of three NK cell populations showing the impact of the DC vaccine-
based control strategy for three different maximum vaccine levels (as shown in the legend).
This plot indicates that increasing the maximum vaccine level increases the peak NK level and
decreases the time to reach that peak.
high at this time. The DCV to be administered during this treatment can be seen
in Figure 3.3. To eliminate the tumour cells as shown in Figure 3.2, the maximum
vaccine value of 2000 units should be administered for approximately the first 58
days, after which the treatment reduces sharply and finally stops at the day 62.
In this simulation, we use parameter B = 0.001.
Increasing the strength of the DC vaccine not only impacts the tumour cell
population, it also causes increases in the numbers of NK cells (see Figure 3.4)
and CD8+ T cells (see Figure 3.5). These effects in turn also reduce the growing
of tumour cell population as shown in Figure 3.6. The effect is very significant in
increasing the number of CD8+ T cells, consequently reducing the tumour burden.
This particular numerical solution shows that DCs alter the population of CD8+
T cells more so than NK cells.
Figures 3.7a, 3.7b and 3.8 display the strength of the DC vaccine to be applied
88
Figure 3.5: The evolution of three CD8+ T cell populations showing the impact of the
DC vaccine-based control strategy for three different maximum vaccine levels (as shown in the
legend). This plot indicates that increasing the maximum vaccine level increases the peak CD8+
T cell level and slightly decreases the time to reach that peak.
89
Figure 3.6: The evolution of three tumour cell populations showing the impact of the DC
vaccine-based control strategy for three different maximum vaccine levels (as shown in the legend).
This plot indicates that increasing the maximum vaccine level decreases the peak tumour cell
level and slightly decreases the time to reach that peak.
90
(a)
(b)
Figure 3.7: The control, u, used for treatment with initial value of tumour cells T0 = 42310
and where the maximum vaccine to be administered is (a) 2000 units and (b) 20000 units.
91
Figure 3.8: The control, u, used for treatment with initial value of tumour cells T0 = 42310
and where the maximum vaccine to be administered is 200000 units.
92
Figure 3.9: The evolution of two tumour cell populations using T0 = 100. Parameter values
taken from Tables 2.1 and 2.2, except for c1 = 1.5 × 10−6. The dashed line shows the tumour
population resulting from no control, while the solid line (more easily seen in the inset plot)
shows the tumour cell population eradication resulting from the optimal control vaccine strategy.
over time given a maximum DC vaccine strength of 2000 units, 20000 units and
200000 units, respectively. Increasing the strength of the DC vaccine also reduces
the required time duration of the treatment.
Similarly, Figure 3.9 represents the evolution of the number of tumour cells us-
ing parameter value taken from Tables 2.1 and 2.2, execpt for c1 = 1.5 × 10−6
and parameter B = 0.9 × 10−11. The dashed line shows the evolution without
control/vaccine and the solid line represents the tumour cell population change
with optimal control application of the vaccine. With initial value T0 = 100 cells,
without vaccine, the tumour cell population grows to a nonzero tumour equilib-
rium after around 80 days. However this can be eliminated in 100 days using the
optimal control vaccine strategy. The inset plot shows the evolution of the tumour
cell populations on a compressed vertical scale. We see more clearly that while the
controlled tumour is eradicated in around 100 days, the tumour cell population
93
(a)
(b)
Figure 3.10: The control, u, used for treatment with an initial value of tumour cells T0 = 100
and where the maximum vaccine to be administered is (a) 2000 units and (b) 20000 units.
Parameter values are taken from Tables 2.1 and 2.2, except for c1 = 1.5× 10−6.
94
without vaccine remains at a high level at this time. The DC vaccine to be admin-
istered during this treatment can be seen in Figure 3.10a. To eliminate the tumour
cells using a maximum strength of 2000 units, the vaccine should be administered
during the overall treatment, whereas for a much higher maximum strength the
time of treatment is reduced significantly (see Figure 3.10b). From this simulation,
it is found that increasing the value of parameter B causes an decreasing in the
number of DC vaccine to be administered.
Figure 3.11: The control, u, used for treatment with an initial population of tumour cells
T0 = 100 and where the maximum vaccine to be administered is 200000 units. Parameter values
are taken from Tables 2.1 and 2.2, except for c1 = 1.5× 10−6.
Again, increasing the strength of the DC vaccine causes an increase in the num-
ber of NK cells and CD8+ T cells (data not shown), however this reduces the
growing of the tumour (see Figure 3.12). The effect is very significant in increas-
ing the number of CD8+ T cells, consequently reducing the tumour burden. This
simulation shows that DCs impact on the CD8+ T cell population more so than
NK cells. Furthermore, Figure (3.10)a, (3.10)b and (3.11) represent the form of the
vaccine strategy to be used with a maximum DC vaccine level of 2000 units, 20000
95
Figure 3.12: The evolution of three tumour cell populations using optimal control with
varying maximum maximum vaccine levels to be administered (as shown on the figure legend).
Parameter values are taken from Table 2.1 and 2.2, except for c1 = 1.5 × 10−6. Intuitively, the
plot indicates that increasing the maximum value for the vaccine decreases both the peak of
tumour cell population and the length of time to eliminate the tumour cells.
96
units and 200000 units respectively. Increasing the strength of the DC vaccine also
reduces the required time duration of the treatment.
3.4 Discussion
In this chapter, optimal control theory has been to determine appropriate dentritic
cell treatments to apply to a growing tumour. Using two conditions which indicate
that the tumour cells grow to the nonzero tumour equilibrium, we have successfully
implemented an optimal control model and been able to find optimal treatment
strategies that result in eradicating tumour cell in optimal time.
The numerical solutions indicate that if the optimal DC treatment is admin-
istered to the growing tumour, then it eliminates the tumour quickly. However,
it should be kept in mind that there is a limitation of DC treatment when it is
applied in the form of an injection to the patient. As mentioned in Mazzolini et
al. [93], treatment using DCs in colorectal cancer was safe and well tolerated, with
injection of up to 5 ×107 DCs.
This model suggests that the best way to control the tumour cell population is
to give a high DC vaccine concentration at the beginning of the treatment and
then reduce the treatment after a specific period of time. This optimal strategy
coincides with the previous work by Castiglione and Piccoli [27]. They found that
to reduce the tumour cells, the high dose of the first vaccination should be given
at the beginning of treatment period, then the smaller doses of vaccination should
be administered later the treatment. This result also supports findings of de Pillis
et al. [40], they found that the best way to fight the growing tumour was to apply
the bulk of chemotherapy at the beginning of the treatment.
In practice however, the scheduling of tumour immunotherapy is periodical. For
example, in a study due to Burgdorf et al., DCs were administered biweekly with
a total of 10 vaccinations per patient [21]. It is however hypothesised that in the
future, DC vaccines can be administered to the patient in what mathematically
would be represented as a continuous function of time, using portable pumps [68].
In future work, we intend to consider a particular case of multitherapy, for which
different factors (DCs and CTL cells or drug treatment) are delivered at the same
time to the patient. As described in de Pillis et al. [37] combined immunotherapy
and chemotherapy are more effective than immunotherapy or chemotherapy alone.
Further developments of the current study include the extension to consider specific
kinds of tumour and also investigation of spatial variations in the development of
the tumour and the effect of the immune system.
Chapter 4
A cellular automata model to investigate immune
cell-tumour cell interactions in growing tumours in
two spatial dimensions
4.1 Introduction
Presented in Chapter 2 was an ODE-based mathematical model of the interactions
between tumour cells and components of the host immune system. Based on this
model, a mathematical model for tumour treatment using an optimally controlled
vaccine application also has been discussed in Chapter 3. In this section, we
present a mathematical model of a growing tumour and the interaction between
the tumour cells and the host immune system using a cellular automata (CA)
model. This model can describe the system in much more detail, including spatial
variations and cell-cell interactions of every single cell in the system. Based on
Mallet and de Pillis [89] model, we construct a 2D CA model instead of 3D model,
since this model can provide a preliminary basis for the project in the future.
As has already been discussed, there is strong evidence in the literature for
the hypothesis that tumour growth is directly influenced by the cellular immune
system of the human host. We stress this point further and note some specific
interactions that, with the cellular automata modelling strategy, we are now able
to incorporate directly into our mathematical model. For example, Hart [65] states
that DCs, found in many types of tumours, are the dominant antigen presenting
cells for initiating and maintaining the host immune response. They are critical
in activating, stimulating and recruiting T lymphocytes: cells with the ability
to lyse tumour cells. Also, Sandel et al. [119] discuss the influence of DCs in
controlling prostate cancer. Furthermore, tumour infiltrating DCs are a key factor
at the interface between the innate and adaptive immune responses in malignant
97
98
diseases. Beside their primary role in the induction and regulation of the adaptive
antitumoural immune response, more recent studies have shown that DCs have
a capacity to directly kill cancer cells [81]. Natural killer cells and cytotoxic T
lymphocyte cells also play important roles in the response of the immune system
against the tumour as described in Kindt et al. [73].
In the tumour micro-environment, the tumour produces chemokines that can
attract immune system components including DCs, NK cells and T cells to sites of
tumour growth. These chemokines also function to activate DCs which, in turn,
can stimulate and activate the immune system [124]. Chemokines are a family
of small cytokines, or proteins secreted by many different cell types, including
tumour cells. They can affect cell-cell interactions and play a fundamental role in
the recruiting or attracting of cells of the immune system to sites of infection or,
of interest in the present research, tumour growth.
As discussed in Chapter 1, the dynamics of tumour growth and the interactions
of growing tumours with the host immune system have been studied intensively us-
ing mathematical models over the past four decades (see also Araujo and McElwain
for a review to the early 2000s). Most of these models are presented using ordi-
nary differential equations (ODEs) or partial differential equations (PDEs) that
impose restrictions on the modelled system’s time-scales, as described in Ribba et
al. [115]. However, a CA model can describe more complex mechanisms in the bi-
ological system without such restrictions by detailing phenomena at the individual
cell or particle level. The classic definition of a CA model imposes that they involve
only local rules that depend on the configuration of the spatial neighbourhood of
each CA element. Hybrid cellular automata (HCA), on the other hand, extend the
CA to incorporate non-local effects, often via coupling the CA with PDEs, and it
is this modelling strategy that we employ in this chapter.
The interactions of a tumour and the host immune system using the CA frame-
work have been modelled previously by, for example, Mallet and de Pillis [89] and
de Pillis et al. [38], where they presented the first multidimensional, hybrid cellular
automata model of the process that incorporated important signalling molecules.
However, these models and others neglected to describe DCs and chemokines and
their roles in tumour growth and control. This present research attempts to im-
prove on their work by explicitly describing more of the host immune system.
The purpose of the model developed in the present research is to investigate the
growth of a small solid tumour when that growth is affected by the immune system.
To this end, in this study, we present a hybrid cellular automata model of the
interaction between a growing tumour and cells of the innate and specific immune
system, including the role of DCs and signalling molecules known as chemokines.
To include the effect of a chemokine in this model, we recognise the significantly
99
smaller size of such molecules compared with biological cells and introduce a par-
tial differential equation to describe the concentration of chemokine secreted by
the tumour. We combine the numerical solution of the partial differential equation
model with a number of biologically motivated automata rules to govern the evo-
lution of various cell populations from the HCA model. We use the hybrid cellular
automata model to simulate the growth of a tumour in a number of computa-
tional “cancer patients”. Each computational patient is distinguished from others
by way of patient-specific characteristics reflected through particular parameter
choices. We define the “death” of a patient as the situation where the cells of
the tumour reach the boundary of our model domain; effectively this represents
tumour metastasis.
In the sections to follow, we present a discussion of the role of DCs and chemokines
related to cell-cell interactions in tumour growth. Furthermore, the development of
the HCA model is considered before analysing numerical simulations. We conclude
with a discussion of the results.
4.2 The role of DCs and chemokines
Before starting to construct the model using the HCA framework, in this section,
we present a more detailed discussion of the role of immune system (especially
DCs) and chemokines as related to cell-cell interactions in tumour growth. This
discussion provides us the basis for making rules in HCA model.
The immune system plays an important role in defending the body against
pathogens by identifying and killing non-self (foreign) matter such as viral parti-
cles, parasites and importantly here, tumour cells [117]. The immune system may
be considered to consist of two components, namely the innate and adaptive sys-
tems. The innate immune system (for example: DCs, NK cells, macrophages) can
recognise antigen without the requirement for previous priming by specific non-self
antigens [117]. However, the adaptive immune system (for example: cytotoxic T
cells, helper T cells and B cells) need the antigen to be processed and presented
in a histocompatibility complex through antigen presenting cells (APC) [11, 117].
DCs are known to be the most efficient APC and express high levels of MHC class
I and II molecules [59].
Current evidence is that there is a large number of functional states for DCs and
the immunogenic capacity depends on the microenvironment [84]. The tumour mi-
croenvironment is a complex system that consists of extracellular matrix [114] and
stromal cells including fibroblasts, endothelial cells, lymphocytes, macrophages,
DCs, and neutrophils, which supports and regulate tumour growth [114, 124].
In the tumour microenvironment, DCs play a crucial role in activating, stimu-
100
lating and recruiting the immune system. DCs can be activated by chemokines se-
creted by tumour cells or after direct interactions with the tumour itself [114, 124].
These activated DCs can stimulate and regulate components of the immune sys-
tem including CTL cells [121] and NK cells, which after activation can directly kill
tumour cells [15, 59]. Furthermore, helper T cells that are activated by DCs [121]
can produce chemokines which, in turn, lead to CTL cell stimulation [124].
Dendritic cells found in various types of solid tumours, are antigen presenting
cells that initiate and regulate immunity as well as shape the host response to
tumours [42, 65]. They play an important role in activating, stimulating, recruit-
ing and developing the immune response. Therefore, it can be concluded that
tumour-infiltrating DCs play a key role in a cellular antitumour immune response
by infiltrating, capturing, and processing tumour antigens, recruiting and activat-
ing the immune system [42, 119].
Dendritic cells play a vital role as the major regulator in CTL cell and NK cell
activation [30, 51, 81], they also can control the activation of B cells [11]. As well,
DCs have a unique function depending on their stage of maturation. Immature DCs
(iDC) are very efficient in antigen uptake and are capable of presenting captured
antigens through their surface receptors. After antigen uptake, DCs migrate from
peripheral tissues to the lymph nodes, where antigen presentation to the immune
system occurs [11, 123]. In the secondary lymphoid tissues, DCs are mature and
able to attract, interact and activate the immune system including T lymphocytes
and helper T cells to initiate a primary immune response [1].
Cytotoxic T cells are activated by DCs through antigens presented to MHC class
I molecules. Also, DCs activate CD4 helper T cells which then secrete chemokines
that can enhance immunoglobulin production. CD4 helper T cells are activated by
binding via their T cell receptor (TCR) to MHC class II molecules. Then they can
be recognized by specific CD4 helper T cells in the cell membrane, where MHC
peptide complex is presented to the CD4 helper T cells by DCs. CD4 cells can be
categorised according to the type of signalling that they receive, Th1 CD4 helper
T cells and Th2 CD4 helper T cells. Th1 CD4 helper T cells secrete chemokines
such as interleukin-2, thereby stimulating cell-mediated immunity by activating
CTL cells. Th2 CD4 helper T cells mediate an antibody response by releasing
chemokines such as IL-4 and IL-10 [59].
On the other hand, active CTL cells can kill mature DCs [26], as a result of
presenting antigen on their surface. In a recent study, cytokine-induced killer T
cells (CIK), expanded T cells from ex vivo have been shown to selectively eliminate
iDC by direct cytotoxicity [69]. Furthermore, Moretta [96] notes that NK cells can
down regulate the function of DCs by killing iDCs in peripheral tissues, and also
states that NK cells might have a role in killing mature DCs.
101
Dendritic cells are considered as the most potent component of the immune sys-
tem because they facilitate transport of antigen-presenting cells to the lymphoid
tissue and provide efficient stimulation of T cells [86]. From experimental studies
in mice, DCs have been shown to be very efficient in stimulating CTL cells [86].
Because of their unique role in initiating and regulating the immune response, cur-
rently DCs are exploited in the hope of becoming a novel tool for cancer therapy. It
has been proven that DCs are feasible, safe [21, 93, 120], and efficient for treatment
of some “cancer patients”, especially if DCs are matured and activated [1]. The
first study related to DC vaccination was carried out by Hsu, et al. in 1996 [66].
They investigated the ability of DCs pulsed ex vivo to stimulate host anti-tumour
immunity in patients with B cell lymphoma. Other research, for example, Schuler
et al. [120] , Fong et al. [58] and Burgdorf et al. [21], also discussed the use of
DCs in cancer immunotherapy.
Chemokines and chemokine receptors, important in immune homeostatis and
surveillance, also play an important role in the tumour environment. In the tu-
mour microenvironment, chemokines are secreted both by stromal cells (fibroblasts,
endothelial cells and infiltrating leukocytes) and by the tumour itself [7, 124]. Tu-
mours, such as glioblastoma, melanoma, and neuroblastoma produce high levels of
chemokines. These chemokines can promote tumour growth and induce stromal
cells to produce cytokines or chemokines which, in turn, can regulate angiogenesis,
tumour growth, and metastasis in the tumour microenvironment [124]. Tumour
cells can activate DCs, lymphocytes, fibroblasts, macrophages and neutrophils,
through cell-cell interactions or cytokines or chemokines produced by tumour cells
[124].
Chemokines secreted by tumour or stromal cells can also attract a large num-
ber of leukocytes such as DCs, NK cells, and T cells (helper T and cytotoxic
T lymphocytes) to the tumour site which may result in tumour regression and
elimination [124]. Chemokines secreted by tumour cells can stimulate or inhibit
tumour growth, modulation of chemokine activity in a selective manner at the
site of the tumour can lead to tumour cell apoptosis. These kinds of chemokines
can be chosen for their potential to attract immune cells to the tumour site that
may result in successful tumour treatment. From experiments involving mouse
tumours, chemokines such as CCL5 and CCL21 are promising avenues for cancer
therapy investigations [102, 127].
4.3 Mathematical model
In this model, we consider the early growth of a solid tumour and its interac-
tion with the immune system and a tumour-secreted chemokine. The model is
102
x
y
CA element
Figure 4.1: Schematic showing the partitioning of the problem domain into cellular automata
elements (squares) and mesh-points for numerical solution of the partial differential equation.
comprised of a system of partial differential equations to describe the chemokines
secreted by the tumour and CD4+ T cells, coupled with a discrete, stochastic cel-
lular automata that describes the various cell types comprising the host-tumour
environment.
Following Ferreira et al. [52, 53] and de Pillis and coworkers [38, 89] the tu-
mour environment is modelled as a square shaped computational domain of side
length L (see Figure 4.1). Each square element in the grid represents a loca-
tion that may contain a healthy cell, tumour cell or immune cell. The domain is
partitioned into a regular square grid with each element of the grid representing
a space approximately corresponding with the size of a tumour cell (around 10-
20µm; [6, 83]). These elements are the discrete locations considered in the cellular
automata component of the model, while the midpoint of each element will be
used as a mesh-point in the numerical scheme used to solve the partial differential
equation component.
Initially, non-cancerous healthy cells cover the whole of the model domain, then
the tumour mass is allowed to grow from one cancer cell placed at the centre cell
of the grid. Cells of the host immune system are initially spread randomly over the
domain throughout the other healthy cells. Four separate immune cell populations
are considered here – the NK cells and DCs of the innnate immune system and
cells of the specific immune response, represented by the CTL cells and helper T
cells. Each of these may exist in either an active or inactive state.
Computationally, the CA grid is stored as a two dimensional data structure
103
Number Description
0 healthy cell
1 tumour cell
2 necrotic debris
3 inactive CD8+ cytotoxic T cell
4 active CD8+ cytotoxic T cell
5 inactive dendritic cell
6 active dendritic cell
7 inactive CD4+ helper T cell
8 active CD4+ helper T cell
9 active NK cell
10 inactive NK cell
Table 4.1: The different cell species tracked in the cellular automata and the numerical value
given to each in the computational implementation.
(matrix) with correspondence between CA elements and matrix elements. The
number stored in the matrix corresponds with the type of cell occupying that
element in the domain, according to the definitions given in Table 4.1.
The model solutions are progressed via discrete time steps, at which each spatial
location is investigated to determine its contents and whether or not certain actions
will occur. This is summarised in Algorithm 1.
Algorithm 1 Brief pseudocode for the full model algorithm.
Draw set parameters for current computational patient
Initialise CA domain contents
Solve PDEs
for each time step do
for each CA element do
Determine cell type in element
Characterise neighbourhood of element
Update PDEs solution (chemokine concentration)
Test whether event will occur and update state
end for
end for
Export data
The rules for the cellular automata component as well as the form of the diffusion
104
equation for chemokines are presented below.
4.3.1 Diffusion equation for chemokine concentration
Chemokines are small (8-14 kDa, in size [114]), cell-secreted protein molecules that
can effect cell-cell interactions. In this model we consider two different chemokine
molecules such as an interleukins. Given that such molecules are very small com-
pared with the size of tumour and host cells, we treat them essentially as a contin-
uum and use a partial differential equation to model changes in their concentration
in space and time. Denoting the concentration of the chemokines as C1(x, y, t) and
C2(x, y, t) we have,
∂C1
∂t= DC1
(∂2C1
∂x2+∂2C1
∂y2
)+ σT, (4.1)
∂C2
∂t= DC2
(∂2C2
∂x2+∂2C2
∂y2
)− λC2 + γ + αDAHI − βC2I, (4.2)
where DC1 and DC2 are the coefficients of random motility of the chemokines.
The parameter σ is the rate of secretion of chemokine by tumour cells. The rate
of chemokine secretion by CD4+ helper T cells is represented by λ as a natural
degradation rate and γ as a natural production rate. The constant α represents
the rate of secretion of chemokine resulting from interactions between activated
DC cells and helper T cells, while β represents the rate at which chemokine is used
up in activating CD8+ cytotoxic T cells. The description of model variables can
be seen in Table 4.2.
Initially and on the boundaries, we assume that there are no chemokines. How-
ever these partial differential equations must be solved at each time step of the
HCA model. Given that chemokines are secreted by the tumour cells and when
CD4+ T helper cells and DC cells come in contact (in the cellular automata com-
ponent of the model), at later times the initial condition used in a HCA time
step becomes nonzero and is in fact provided by the outcomes of the cell level
interactions.
4.3.2 Cellular automata rules
In this model, we consider a number of biological cell types including normal
healthy cells, tumour cells (necrotic, dividing and migrating), DCs, NK cells, CTL
cells and helper T cells. To build the CA model, we define ‘rules’ that draw
upon the biological literature to describe cell-cell interactions, cell effects on the
environment, and effects of the environment on cells.
The evolution of the cell species involved in the tumour-host interactions con-
sidered here is governed by a set of discrete, stochastic rules which are presented
105
Variable Description
C1(x, y, t) concentration of chemokine secreted by tumour
cells
C2(x, y, t) concentration of chemokine secreted by CD4+T
cells
Ni,j healthy host cell
Ti,j tumour cell
DIi,j inactive dendritic cell
DAi,j active dendritic cell
IIi,j inactive CD8+ cytotoxic T cell
IAi,j active CD8+ cytotoxic T cell
HIi,j inactive CD4+ helper T cell
HAi,j active CD4+ helper T cell
KIi,j inactive NK cell
KAi,j active NK cell
Table 4.2: The variables used in the hybrid cellular automata model. Here x and y are
the spatial variables for the PDE component, t denotes the time variable, and i and j represent
spatial locations in the cellular automata component.
106
below. Each particular cell-level action or interaction has associated with it, a
probability of success. Generally speaking, we calculate the number
P cellevent = f(.),
where f depends on relevant cell types and conditions in the neighbourhood . We
compare P cellevent with a pseudo-random number, r, drawn from the uniform distri-
bution on the interval [0,1]. If r < P cellevent then the event is carried out, otherwise
it is deemed to have failed to occur. To describe the evolution of the cell popula-
tion, we introduce the general algorithm of cellular automata rules as presented in
Algorithm 2.
Algorithm 2 Pseudocode for testing occurrence of individual events.
Draw r ∼ U [0, 1]
Calculate P cellevent using current state of CA
if r < P cellevent then
update state (the event occurs)
end if
We now consider each cell type in turn and introduce the specific forms of the
CA rules utilised in the model.
Host cells
Following from the work of Ferreira et al. [52] and of Mallet and de Pillis [89],
we assume that the healthy host cells are effectively passive bystanders in the
interaction. They do not hinder the growth of the tumour cells or the movement
of any cell type.
Tumour cells
In this model, we consider tumour growth to be influenced by the immune system
via NK cells, CTL cells and DCs. The tumour cells undergo the processes of
division, migration and lysis resulting from interactions with components of the
immune system such as NK cells, CD8+ cytotoxic T cells, and DCs. We assume
lysis is dependent upon the local strength of the immune system and model division
to be influenced by crowding due to the presence of other tumour cells, respectively.
At each time step, the neighbourhood of each tumour cell is surveyed to determine
whether the cells of the active immune system are present or not. If immune
cells are present, the tumour cell is marked for potential lysis whereas if there
are no active immune system cells in the neighbourhood then the tumour cell is
107
marked for potential division or migration. A stochastic rule is then implemented
to determine whether or not the action (division, migration or lysis) will be carried
out. While the time-scales of these processes can vary (between or within cell
types), for generality we consider equal time-scales and tie to this the time step
of the numerical solution method. With this in mind, we impose the following
cellular automata rules for tumour cells.
Cell division: For cell division, we adopt a probability of division of the similar
form as that used by Mallet and de Pillis [89]. When a tumour cell is marked
for division, that action is carried out with a probability that depends upon the
density of tumour cells in the neighbourhood of the dividing cell. In particular, we
have
PTdiv = exp
−(θdiv
∑i,j∈η
Ti,j
)2,
where θdiv controls the shape of the curve allowing it to capture qualitative under-
standing of the biology and∑
i,j∈η Ti,j is the number of tumour cells in a one cell
radius of the cell of interest.
When division occurs, the resulting daughter cell is placed in an element of
the neighbourhood of the dividing cell in the following order: filling an empty
element, replacing (killing and consuming/removing) a healthy host cell, adding
to the tumour burden of the least filled neighbouring element. From Figure 4.2(a),
it can be seen that tumour cell division is more likely when there is space in the
neighbourhood for the resulting daughter cell.
Lysis by active immune cells: Again for tumour lysis, we adopt a probability of
lysis of the similar form as that used by Mallet and de Pillis [89]. Active immune
cells such as CD8+ T cells, NK cells and DCs are able to directly lyse tumour
cells when they share a local neighbourhood. We assume that the intensity of the
immune system effect is proportional to the number of active CD8+ T cells, NK
cells and DCs in the neighbourhood of the tumour cell. The probability of tumour
lysis depends on the strength of the active immune system in the neighbourhood
of the tumour cell (see Figure 4.2(b)), and is given by
PTlysis = 1− exp
−(θlysis
∑i,j∈η
(IAi,j +KA
i,j +DAi,j
))2,
where again θlysis controls the shape of the curve allowing it to capture qualitative
understanding of the biology and∑i,j∈η
(IAi,j +KA
i,j +DAi,j
),
is the number of active immune cells in a one cell radius of the tumour cell of
interest.
108
Cell migration: At each time step the tumour cells marked for migration do so
with a constant probability, k1 as given below
PTmig = k1.
These rules are presented in pseudocode form in Algorithm 3.
Algorithm 3 Pseudocode for tumour cell related events.
if (location holds tumour cell) then
Calculate the number of tumour cells in the neighbourhood
Calculate the number of immune cells in the neighbourhood
Find new location at random in the neighbourhood
Draw r ∼ U [0, 1]
Calculate PTdiv using current state of CA
Calculate PTlysis using current state of CA
if r < PTdiv and (new location holds healthy cell) then
filling new location by tumour cell
else if r < PTlysis then
current surface become necrotic debris
else if r < PTmig then
cell move
end if
end if
CD4+ helper T cells
Inactive CD4+ helper T cells are subject to change as a result of natural replenish-
ment and activation due to direct interaction between dendritic cells and existing
inactive CD4+ helper T cells. At each time step, the neighbourhood of each inac-
tive CD4+ helper T cell is surveyed to determine whether DCs are present. If any
DCs are present, then the CD4+ helper T cell is marked for potential activation,
otherwise the CD4+ helper T cell is marked for random migration. A stochastic
rule is then implemented to determine whether or not the action (migration or ac-
tivation) will be carried out. A normal background level of inactive CD4+ helper
T cells is also maintained at each time step. To this end, we impose the following
cellular automata rules.
Activation following DC and inactive CD4+ helper T cell contact: When inactive
helper T cells come in contact with DCs they have probability
PCD4act = 1− exp
(−∑i,j∈η
(DAi,j
)2
),
109
(a)
(b)
Figure 4.2: The form of the curves used to determine the probability of (a) tumour cell
division and (b) tumour cell lysis, given different neighbourhood conditions.
110
of becoming active CD4+ T cells which are then able to secrete chemokines to
activate cytotoxic T cells (see equation (4.2)). Here∑i,j∈η
(DAi,j
),
is the number of active DC cells in a one cell radius of the cell of interest.
CD4+ T cell migration: If any DCs are present, then the CD4+ helper T cell is
marked for potential activation, otherwise the CD4+ helper T cell is marked for
random migration with probability of migration given by
PCD4mig = k2.
Natural replenishment to background level: A near-constant minimum background
level of inactive and active helper T cells is ensured at each time step by replacing
some healthy cells on the boundary with inactive helper T cells. This mimics
replenishment of the CD4+ population from external sources (such as the lymph
nodes). At each time step we determine the proportion of all locations in the
domain occupied by inactive and active helper T cells. Whenever this is less than
the minimum background level, H0, each healthy cell on the boundary of the
domain has probability
PCD4rep = H0 −
1
n2
∑i,j∈η
(HIi,j +HA
i,j
),
of being replaced with an inactive CD4+ T cell from outside of the problem domain,
where H0 is the ‘normal’ density of inactive CD4+T helper cells and n2 is the total
number of CA elements.
These rules are presented in pseudocode form in Algorithm 4.
CD8+ cytotoxic T cells
Inactive CD8+ cytotoxic T cells are subject to change as a result of natural replen-
ishment and activation as a result of direct interaction between existing inactive
CD8+ cytotoxic T cells and either active DCs, tumour cells or cytokines produced
by active CD4+ helper T cells. At each time step, the neighbourhood of each inac-
tive CD8+ cytotoxic T cell is surveyed to determine whether active DCs or tumour
cells are present. If any DCs or tumour cells or cytokines are present, then the inac-
tive CD8+ cytotoxic T cell is marked for potential activation, otherwise the CD8+
cytotoxic T cell is marked for random migration or movement towards regions of
higher chemokine concentration. If the chemokine level secreted by active CD4+
T cells is greater than some threshold concentration, C20 , then the inactive CD8+
cytotoxic T cell is marked for activation. A stochastic rule is then implemented to
111
Algorithm 4 Pseudocode for CD4+ helper T cell related events.
if (location holds inactive CD4+ helper T cell ) then
Calculate the number of tumour cells in the neighbourhood
Find active DCs in the neighbourhood
Calculate the number of active DCs in the neighbourhood
Draw r ∼ U [0, 1]
Calculate PCD4act using current state of CA
if The number of tumour cells in the neighbourhood >= 1 then
current surface replace with healthy cell
else if r < PCD4act then
current surface replace with active CD4+ helper T
else if r < constant then
inactive CD4+ helper T cell moves towards the higher chemokine concen-
tration
end if
end if
determine whether or not the action (migration or activation) will be carried out.
A normal background level of inactive CD8+ T cells is also maintained at each
time step. As a result, we impose the following cellular automata rules.
Activation following DC and/or tumour cell and inactive CD8+ T cell contact:
When inactive cytotoxic T cells come in contact with DCs and/or tumour cells
they have probability
PCD8act = 1− exp
−(∑i,j∈η
DAi,j
)2,
of becoming active CD8+ T cells, which are then able to lyse tumour cells. Here∑i,j∈η
(DAi,j
),
is the number of active DC cells in a one cell radius of the cell of interest. We
assume that active CD8+ cytotoxic T cells can lyse tumour cells more than once.
CD8+ cytotoxic T cells are neutralised if there are no more tumour cells in the
neighbourhood. CD8+ cytotoxic T cells can also kill active DCs. At each time step,
the neighbourhood of each active CD8+ cytotoxic T cell is surveyed to determine
whether active DCs are present. If any DCs are present, then the CD8+ cytotoxic
T cells lyse the DCs.
CD8+ T cell migration: At each time step the inactive CD8+ T cells marked for
migration do so with a constant probability given by,
PCD8mig = k3.
112
Natural replenishment to background level: Similarly to the helper T cells, a near-
constant minimum background level of inactive and active cytotoxic T cells is
maintained at each time step by replacing some healthy cells on the boundary
with inactive cytotoxic T cells. At each time step we determine the proportion
of all locations in the domain occupied by inactive and active cytotoxic T cells.
Whenever this is less than the minimum background level, I0, each healthy cell on
the boundary of the domain has probability
PCD8rep = I0 −
1
n2
∑i,j∈η
(IIi,j + IAi,j
),
of being replaced with an inactive CD8+ T cell from outside of the problem domain,
where I0 is the ‘normal’ density of inactive CD8+ T cells and n2 is the total number
of CA elements.
These rules are presented in pseudocode form in Algorithm 5.
Algorithm 5 Pseudocode for inactive CD8+ T cell related events.
if (location holds CD8+ T cell) then
Calculate the number of tumour cells in the neighbourhood
Find active DCs in the neighbourhood
Calculate chemokine concentration in the neighbourhood
Find new location at random in the neighbourhood
Draw r ∼ U [0, 1]
Calculate PCD8act using current state of CA
if the number of tumour cells in the neighbourhood >= 1 then
current surface replace with active CD8+ T cell
else if r < PCD8act then
current surface replace with active CD8+ T cell
else if chemokine concentration > threshold chemokine concentration then
current surface replace with active CD8+ T cell
else if r < k3 then
inactive CD8+ T cells move to the new location
end if
end if
Dendritic cells
Inactive DCs are activated when they come in contact with either chemo-kines
secreted by tumour cells or with the tumour itself. DCs process the tumour asso-
ciated antigens and present the antigen on their cell surface. Active DCs play an
important role in the activation of T cells and can also be lysed by activated CD8+
113
cytotoxic T cells as a result of presenting antigen on their surface. At each time
step, the neighbourhood of each inactive DC is surveyed to determine whether
tumour cells or chemokines are present nearby. If either are present, the DCs are
marked for potential activation. The neighbourhood of each active dendritic cell is
surveyed for the presence of active CD8+ T cells. If the chemokine concentration
level secreted by tumour cells is greater than some threshold concentration, C10 ,
then the inactive DC is marked for activation. When active cytotoxic T cells reside
nearby, the DC is marked for potential lysis. A stochastic rule is then implemented
to determine whether or not the action (migration, lysis or activation) will be car-
ried out. A normal background level of inactive DCs is also maintained at each
time step. As a result, we impose the following cellular automata rules.
Activation by interaction with chemokines and/or tumour cells: Inactive DCs pro-
cess and present tumour associated antigen upon interaction with tumour cells.
That is, active DCs are activated with a probability given by
PDCact = 1− exp
−(θact
∑i,j∈η
Ti,j
)2,
where θact controls the shape of the curve allowing it to capture qualitative under-
standing of the biology and∑
i,j∈η Ti,j is the number of tumour cells in a one cell
radius of the cell of interest.
Lysis by active CD8+ T cells: CD8+ T cells kill DCs presenting antigen with
probability
PDClysis = k4.
Dendritic cell migration: At each time step the inactive DCs marked for migration
do so with a constant probability
PDCmig = k5.
Natural replenishment to background level: Similarly, a near-constant minimum
background level of inactive and active DCs is maintained at each time step by
replacing some healthy cells on the boundary with inactive DCs. At each time step
we determine the proportion of all locations in the domain occupied by inactive
and active DCs. Whenever this is less than the minimum background level, D0,
each healthy cell on the boundary of the domain has probability
P inactDCrep = D0 −
1
n2
∑i,j∈η
(DIi,j +DA
i,j
), (4.3)
of being replaced with an inactive DC from outside of the problem domain, where
D0 is the ‘normal’ density of inactive DCs and n2 is the total number of CA
elements.
These rules are presented in pseudocode form in Algorithm 6.
114
Algorithm 6 Pseudocode for DCs related events.
if (location holds inactive DCs) then
Calculate the number of tumour cells in the neighbourhood
Calculate chemokines concentration in the neighbourhood
Find new location at random in the neighbourhood
Draw r ∼ U [0, 1]
Calculate PDCact using current state of CA
if number of tumour cells in the neighbourhood >= 1 then
if r < PDCact then
current surface replace with active DCs
end if
else if chemokine concentration > threshold chemokine concentration then
current surface replace with active DCs
else if r < k5 then
inactive DCs move to the new location
end if
end if
Natural killer cells
Inactive NK cells are subject to change as a result of natural replenishment and
activation as a result of direct interaction between existing inactive NK cells and
either active DCs, tumour cells or cytokines produced by tumour cells. At each
time step, the neighbourhood of each inactive NK cell is surveyed to determine
whether active DCs or tumour cells are present. If any DCs or tumour cells or
cytokines are present, then the inactive NK cell is marked for potential activation,
otherwise the NK cell is marked for random migration. If the chemokine level
secreted by tumour cells is greater than some threshold concentration, C10 , then
the inactive NK cell is marked for activation. Active NK cells will survey their
neighbourhood to determine whether tumour cells are present. If any tumour
cells are present, then the active NK cell is marked as having potential to lyse
tumour cells. A stochastic rule is then implemented to determine whether or not
the action will be carried out. A normal background level of inactive NK cells is
also maintained at each time step. To this end, we impose the following cellular
automata rules.
Activation by interaction with DCs: When inactive NK cells come in contact with
DCs, they have probability
PNKact = 1− exp
−(∑i,j∈η
DAi,j
)2,
115
of becoming active NK cells, which are then able to lyse tumour cells. Here∑i,j∈η
DAi,j,
is the number of active DCs in a one cell radius of the cell of interest.
NK cells migration: At each time step the inactive NK cells marked for migration
do so with a constant probability given by,
PNKmig = k6,
where k6 is a constant.
Natural replenishment to background level: Similarly, a near-constant minimum
background level of inactive and active NK cells are maintained at each time step
by replacing some healthy cells on the boundary with inactive NK cells. At each
time step we determine the proportion of all locations in the domain occupied by
inactive and active NK cells. Whenever this is less than the minimum background
level, K0, each healthy cell on the boundary of the domain has probability
P inactNKrep = K0 −
1
n2
∑i,j∈η
(KIi,j +KA
i,j
),
of being replaced with an inactive NK cell from outside of the problem domain,
where K0 is the ‘normal’ density of inactive NK cells and n2 is the total number
of CA elements.
These rules are presented in pseudocode form in Algorithm 7.
4.4 Simulation and Results
The numerical simulation of the model involves two main steps. Spatial changes for
the chemokine species are determined by solving the partial differential equation
using an explicit forward in time, central in space (FTCS) method [18]. Then the
cell-level phenomena (such as cell-cell interactions, cell death, division and migra-
tion) are carried out, dependent on the updated chemokine levels, by updating the
cellular automata component of the model. We tie the time step of this iteration
process to approximate period of tumour cell division (approximately 0.5-10 days;
see for example, [77, 116]).
We combine the solution of the PDE with the CA as described in Section 4.3.2
to simulate the evolution of the growing tumour. Here, a two-dimensional regular
100× 100 square domain is used with 100 cell cycles and a Moore neighbourhood
is considered for the cellular automata rules. In this simulation, we solve the
PDE model using the finite difference method and an estimated value of diffusion
116
(a) (b)
(c) (d)
Figure 4.3: The growing tumour and host immune system. After 25 cell cycles (a). After 50
cell cycles (b). After 75 cell cycles (c). After 100 cell cycles (d).
117
Algorithm 7 Pseudocode of NK cell related events.
if (location holds inactive NK cell) then
Calculate the number of tumour cells in the neighbourhood
Find active DCs in the neighbourhood
Calculate chemokine concentration in the neighbourhood
Find new location at random in the neighbourhood
Draw r ∼ U [0, 1]
Calculate PNKact using current state of CA
if the number of tumour cells in the neighbourhood >= 1 then
current surface replace with active NK cell
else if r < PNKact then
current surface replace with active NK cell
else if chemokine concentration > threshold chemokine concentration then
current surface replace with active NK cell
else if r < k6 then
inactive NK cells move to the new location
end if
end if
coefficient for chemokine, DC1 and DC2 , is 10−4 µm2s−1. The distribution of the
growing tumour after 25, 50, 75 and 100 cell cycles is shown in Figure 4.3, with
results qualitatively matching those of Mallet and de Pillis [89] and commonly
found in tumour modelling literature.
Figure 4.4 shows the evolution of the tumour cell and necrotic cell densities over
100 cell cycles. This plot shows the characteristic exponential and linear growth
phases of solid, avascular tumours (see for example, Folkman and Hochberg [57]), as
well as a slower growing population of necrotic cells. Figure 4.5 shows the number
of tumour cells for 100 simulations (thin lines) and the median simulation (thick
line) of the CA model over 100 cell cycles. In addition, increasing the number of
immature DCs in the domain causes a decrease the number of tumour cells (this
result is run by 100 simulations), see Figure 4.6. This result strongly supports the
hypothesis that DCs can be used as a cancer treatment.
In Figure 4.7(a) we see that initially, the number of mature DCs is zero until
immature DCs come in contact with tumour cells, at which point the matured
DCs commence killing the tumour cells. Also, immature DCs are activated by
chemokines secreted by the tumour cells. As expected, due to the nature of equa-
tion (4.3), the populations of immature and mature DC cells remain approximately
steady over the extent of the tumour growth. Similarly, the same behaviour occurs
in the populations of CTL cells and helper T cells, see Figure 4.7 and Figure 4.8.
118
Figure 4.4: Total cell counts of tumour and necrotic cells after 100 cell cycles. This plot
shows a slower growing population of necrotic cells and an exponential growth of tumour cells
119
Figure 4.5: Total cell counts of tumour cells for 100 simulations (thin) and the median
simulation (thick) of CA model over 100 cell cycles.
120
Figure 4.6: Total cell counts of tumour cells for 100 simulations showing the effect of varying
the DCs as indicated on the graph over 100 cell cycles. This plot shows that increasing the
population of DCs decreases the number of tumour cells.
121
(a)
(b)
Figure 4.7: Total cell counts of CD8+ T cells (a) and DCs (b), after 100 cell cycles.
122
(a)
(b)
Figure 4.8: Total cell counts of CD4+ helper T cells (a) and NK cells (b), after 100 cell
cycles.
123
0 100 200 3000
20
40
60 aCD8+
iCD8+
(a) CTL cells
0 100 200 3000
10
20aDCiDC
(b) Dendritic cells
0 100 200 3000
10
20
30 aTHiTH
(c) T helper cells
0 100 200 3000
5
10aNKiNK
(d) Natural killer cells
0 100 200 3000
0.5
1
·104
HealthyTumourNecrotic
(e) Healthy, tumour and necrotic cells
Figure 4.9: The evolution of CD8+ T cells, Dendritic cells, T Helper cells, NK cells, healthy
cells, tumour cells and necrotic cells for 5 simulations over 300 cell cycles. Parameters used are:
I0 = 0.005, D0 = 0.002, H0 = 0.002,K0 = 0.001, Pdiv = 0.5, Pmig = 0.2
124
0 100 200 3000
50
100 aCD8+
iCD8+
(a) CTL cells
0 100 200 3000
10
20aDCiDC
(b) Dendritic cells
0 100 200 3000
10
20
30
aTHiTH
(c) T helper cells
0 100 200 3000
5
10aNKiNK
(d) Natural killer cells
0 100 200 3000
0.5
1
·104
HealthyTumourNecrotic
(e) Healthy, tumour and necrotic cells
Figure 4.10: The evolution of CD8+ T cells, Dendritic cells, T Helper cells, NK cells, healthy
cells, tumour cells and necrotic cells for 5 simulations over 300 cell cycles. Parameters used are:
I0 = 0.009, D0 = 0.002, H0 = 0.002,K0 = 0.001, Pdiv = 0.9, Pmig = 0.2
125
Figure 4.9 represents the evolution of cytotoxic T cells, DCs, T Helper cells,
NK cells, healthy cells, tumour cells and necrotic cells for 5 simulations over 300
cell cycles. Parameters used are: I0 = 0.005, D0 = 0.002, H0 = 0.002, K0 =
0.001, Pdiv = 0.5, Pmig = 0.2. The initial number of CD8+ T cells is approximately
50 cells or 0.005 % of the total cells in the domain (see Figure 4.9 (a)). In this case,
the tumour cells reach the boundary after approximately 140 cell cycles (see Figure
4.9 (e)). However, increasing the probability of tumour cell division, Pdiv = 0.9,
the tumour cells reach the boundary only after approximately 100 cell cycles (see
Figure 4.10 (e)). In this case, we also increase the initial value of CD8+ T cells. It
means that the probability of tumour cell division is more dominant to affect the
growth of the tumour than increasing immune cells.
We also use the hybrid cellular automata model to investigate the growth of a
tumour in a number of computational “cancer patients”. Each computational pa-
tient is distinguished from others by altering model parameters. We define “death”
of a patient as occurring when the tumour is able to metastasise. Effectively, this
is when the cells of the tumour reach the boundary of our model domain. We
define Kaplan-Meier “survival” estimates as
S(t) =number of individuals “surviving” longer than t
total number of individuals studied,
where t is a time from initial diagnosis to “death”. We present the results of these
simulations using a simulated Kaplan-Meier survival curve, shown in Figure 4.11
and Figure 4.12.
Figure 4.11 describes that metastasis sets in for the first patients after approxi-
mately 80 cell cycles. In addition, at 300 cell cycles after one tumour cell is allowed
to grow, the metastasis of the simulated tumours occurred in approximately 30%
of patients with lowest iDC in the domain, but approximately 50% of the patients
with the highest iDC in the domain. This means that increasing the number of
immature DC in the domain result in significantly longer “survival”. These results
qualitatively agree with experimental data, see for example, Becker et al., Daud et
al. and Nagorsen et al. [13, 32, 100].
Similarly, as shown in Figure 4.12, the patients showed better “survival” as the
number of inactive CTLs within the domain increases. These results also agree
with experimental data as explained in [101].
4.5 Discussion
In this chapter, we have developed a hybrid cellular automata model to describe the
interaction between a growing tumour and the immune system including chemokines.
The model is able to describe the effect of the immune system and chemokines on a
126
Figure 4.11: Simulated Kaplan-Meier curve with different initial values of immature DCs
as indicated on the graph. This plot shows that increasing the population percentage of DCs in
patients will increase “survival” rate.
127
Figure 4.12: Simulated Kaplan-Meier curve with different initial values of immature CTL
as indicated on the graph. This plot shows that increasing the population percentage of CTL
cells in patients will increase “survival” rate.
128
growing tumour. Increasing the number of immature DCs in the domain causes a
decrease in the number of tumour cells. This result strongly supports the hypoth-
esis that DCs can be used as a cancer treatment. Furthermore, we also use the
hybrid cellular automata model to investigate the growth of a tumour in a number
of computational “cancer patients”. Using these virtual patients, the model can
explain that increasing the number of DCs in the domain causes longer “survival”.
Not surprisingly, the model also reflects the fact that the parameter related to
tumour division rate plays an important role in tumour metastasis.
In previous work, Duchting and Vogelsaenger [47] pioneered the use of discrete
cellular automata for modelling cancer, in an investigation of the effects of radio-
therapy. Ferreira et al. [52] modelled avascular cancer growth with a CA model
based on the fundamental biological processes of proliferation, motility, and death,
including competition for diffusing nutrients among normal and cancer cells. Mal-
let and de Pillis [89] constructed a hybrid cellular automata cancer model that
built on the work of Ferreira et al. to include NK cells as the innate immune sys-
tem and CTL cells as the specific immune system. The Mallet and de Pillis model
was lacking in its detail of the immune system and in this present research we have
improved on their work by explicitly describing more of the host immune system.
While direct comparison of the models is difficult, the results as presented in ths
chapter qualitatively reflect the findings of Mallet and de Pillis and of Ferreira et
al. while extending them to incorporate greater realism in the description of the
immune system.
While models based on differential equations allow for analytical investigations
such as stability and parameter sensitivity analyses, and ease of fitting the model
to experimental data, these types of models cannot capture the detailed cellular
and sub-cellular level complexity of the biological system. On the other hand,
HCA models can describe in far greater detail, the intricacies of the biological
process such as the interaction between individual cells. In current work comple-
mentary to the present research of this chapter, we have included greater realism in
the modelling of tumour-secreted chemokines by allowing secretion due to cell-cell
interaction. Currently, chemokines and their receptors in the tumour microenvi-
ronment are being extensively investigated to produce therapeutic interventions
to combat cancer, (see for example, Allavena et al. [7] and Murooka et al. [95]).
Future developments based upon this model will allow for simulation-based and
theoretical investigations of such interventions.
In this chapter, we have developed a useful model that can be employed as a
preliminary investigative tool for experimentalists who conduct expensive in vitro
and in vivo experiments to test and refine hypotheses prior to entering the lab.
With further cross disciplinary collaboration, this type of model can be refined to
129
provide a more accurate description of the underlying cancer biology and hence
yield more relevant predictions and tests of hypotheses.
Chapter 5
Conclusions and Discussion
5.1 Summary of thesis aims
This section summarises the achievement of all aims given in Chapter 1, which are
restated here as follows.
1. To develop a differential equation-based mathematical model of a growing tu-
mour, host tissue and immune system, the associated interactions and result-
ing outcomes. Based on de Pillis and Radunskaya model [34] and Castiglione
and Piccoli [26], we construct new mathematical models that explain more
detail the role of DCs incorporating with NK cells, CD8+ T cells and tumour
cells. These models can provide a tool to describe qualitative relationships
based on particular laboratory findings.
2. To develop a mathematical model of DC-based immunotherapeutic treat-
ment of a growing tumour, that can provide a basic level of understanding
and feedback to experimentalists to assist in designing more effective and/or
efficient laboratory experiments.
3. To develop a cellular automata model of tumour-immune system interac-
tions that explicitly accounts for cell-cell interactions, incorporates a multi-
dimensional spatial viewpoint and allows for stochastic variation to simulate
virtual patients. Based on Mallet and de Pillis model [89], we build a new
cellular automata madel that describe more detailing in immune system. The
effect of chemokines in a growing tumour and the simulation of virtual pa-
tients using similar Kaplan-Meier curve also provide a new contribution of
the literature. This model then provide a deeper level of understanding of
the immune system interactions with a growing tumour.
130
131
Achievement of aim 1
We have developed two new ODE models describing the interaction between a
growing tumour and cells of the innate and specific immune system as presented
in Chapter 2. These models consist of four populations, namely, tumour cells and
three components of the immune system: NK cells, DCs and CD8+ T cells. Under
certain conditions, the first model has a drawback, that is, the evolution NK cells
will grow uncontrolled. To overcome this problem, the model is revised to produce
a new more detailed model by including a saturation effect for a number of the
dynamic processes modelled via Michaelis-Menten kinetics. Also, the recruitment
of DCs which is affected by the number of tumour cells is added in the second
model.
Furthermore, we analysed the stability of the models as well as the simple bifur-
cation behaviour. Both of these models exhibit similar stability behaviour around
their system equilibria. As long as the source terms for NK cells and DCs are
nonzero, there is no trivial equilibrium. If there is no constant source term, that
is both of the source terms for NK cells and DCs are zero, there is a trivial equi-
librium which is always unstable. There is a critical initial value for tumour cells,
where below this critical initial value, the tumour cells can be eliminated, whereas
above this critical initial value, the tumour cells always grow to the nonzero tu-
mour equilibrium. We also analysed the bifurcation behaviour. Under certain
conditions, tumour cells can be eliminated as given in Figure 2.2 and 2.9, on the
other hand, when these conditions are not satisfied the nonzero tumour equilib-
rium is reached. This analysis illustrates the important parameters in achieving a
tumour-free equilibrium, such as the growth rate of tumour cells, the rate at which
tumour cells are killed by NK cells and the rate at which DCs lyse tumour cells.
Finally, numerical solutions of the model demonstrated that the role of DCs
as antigen presenting cells was important in enhancing the immune system. By
increasing the source term for DCs, the population of NK cells and CD8+ T cells
was increased which in turn decreases the number of tumour cells. These results
suggested that an external source of DCs might be used as a treatment for a patient
who lacks sufficient DCs in their body. From these simulations, it is shown that
DCs alone are not fully effective for tumour treatment. It is also necessary that the
NK cell population is high enough for an effective antitumour response. On the
other hand, increasing the source term of NK cells causes decreases to the number
of DCs and CD8+ T cells, suggesting that according to this model NK cells are not
useful as a treatment strategy when compared with DCs. Based on these models,
we undertook further investigation by extending the model to include a DCV as a
tumour therapy strategy.
132
Achievement of aim 2
Presented in Chapter 3 is a mathematical model to describe the dynamics of the
interaction of a growing tumour and the host immune system including a DCV as
a tumour treatment. The second model in Chapter 2 is extended to include the
DCV. Optimal control theory is then applied to this model to find the optimal
choice for the timecourse of administration of DCs to the tumour site, with a
view to minimising the tumour burden. In this problem, we are interested in the
solution of the model in Chapter 2 where the growing tumour tends to the nonzero
tumour equilibrium state.
To provide an overview to the reader, before solving the problem, we derive the
necessary condition for the existence of an optimal control for such a problem.
Based on these conditions, we prove the existence of the optimal control for this
particular problem and then use the Hamiltonian function to solve the objective
functional which has the aim to minimise the number of tumour cells as well as
the cost of the DCV to be administered.
The forward-backward sweep method, which is based on an order 4 Runge-
Kutta scheme, is used to numerically solve the optimal control problem. First,
a forward method is used to solve the state system with its associated initial
condition, then a backward method is applied to solve the adjoint system with
an initial condition obtained from the current iteration of the state system, to
meet the transversality condition. Numerical solution of the model provided the
optimal strategy by which DCs should be administered to the tumour site while
minimising the tumour burden. While the solutions obtained here are of a different
form from strategies commonly used in clinical practice, we note that it has been
hypothesised that portable pumps (not yet developed for clinical use) could be
used to administer a DCV to the patient as a continuous function of the form
found in Chapter 4.
Achievement of aim 3
Chapter 4 of the thesis included the development of a mathematical model of
tumour-immune system interactions that explicitly accounts for cell-cell interac-
tions and spatial variations. The model developed is a two dimensional hybrid
cellular automata for tumour-immune system interactions. Besides providing more
detail of the host immune system than either previous HCA models or the ODE
models developed in this thesis, this model also include chemokines which function
to activate DCs. To include the effect of chemokines in the model, we introduce a
partial differential equation to describe the concentration of chemokine, and couple
this with the discrete cellular automata used to model individual cells.
133
The cellular automata is built as a square shaped computational domain. Each
square element in the grid represents a location that may contain a healthy cell,
tumour cell or immune cell. Initially, one cancer cell is placed at the centre cell of
the grid and non-cancerous healthy cells cover the remainder of the model domain.
A stochastic rule is then applied to each cell to possibly change its current state
due to the state of its neighbours. At each time step, a normal background level
of inactive immune cells is also maintained.
The model described the evolution of tumour cells both in space and time. We
also use the hybrid cellular automata model to investigate the growth of a tumour
in a number of computational “cancer patients”. Each computational patient is
distinguished from others by altering model parameters. We define “death” of a
patient as occurring when the tumour is able to metastasise. Effectively, this is
when the cells of the tumour reach the boundary of our model domain. From
simulation, the model indicated that increasing the number of immature DCs in
the domain result in significantly longer “survival”. This model is able to describe
the complex mechanisms of the interactions between a growing tumour, immune
system and chemokines in individual cells. In future work this model could be
extended to incorporate a discrete analogue of the optimal control theory used to
analyse vaccine treatment in the ODE models developed in this thesis.
The ODE models describe tumours of about 107 cells whereas the CA model
only describes tumours of at most 104 cells. Due to this, care should be taken
in making any comparisons between the work in Chapter 2 and 3 and that of
Chapter 4. In order to increase the size of the tumour modelled using the CA
method would require vast increases to the size of the CA grid and hence also to
the computational power and programming complexity required. Such work was
beyond the scope of this project.
5.2 Contribution of this thesis
The interactions between a growing tumour and the immune system with or with-
out immunotherapy are still somewhat poorly understood from a biological and
immunological point of view. Therefore, the models developed in this research will
make a significant contribution by providing a level of theoretical understanding
of these interactions. The significant impacts of this research include
• providing a tool to describe qualitative relationships based on particular lab-
oratory experiments/findings,
• providing a basic level of theoretical understanding to experimentalists, with
an aim to assisting in designing more effective and efficient laboratory exper-
134
iments related to tumour growth and resulting interactions with the immune
system and immunotherapy,
• developing a deeper level of understanding of the immune system interactions
with a growing tumour.
The ODE models given in this thesis contribute to the theoretical modelling
literature especially for the interaction between a growing tumour and immune
system. The models can provide information of use in designing laboratory ex-
periments, regarding which parameters play an important role in eliminating the
tumour cells. With further data from subsequent experiments, the models can
easily be fit to the data, then further analysis could be conducted. The models are
also able to explain how DCs can alter the growing tumour and other elements of
the immune system. While there is a significant body of literature that describes
a growing tumour and the host immune system, few of these describe the growing
tumour and its interaction with DCs as well as a tumour treatment, and none in
the way shown in this thesis.
In addition, by using optimal control theory the ODE models developed in Chap-
ter 2 have been extended to determine optimal strategies for administering DCV
to the patient in other to minimise the cost of the vaccine as well as the tumour
burden. This model suggests that the optimal treatment strategy, namely to first
administer a high level DCV at the beginning of the treatment schedule with
a subsequent reduced treatment level, actually quite different from the standard
methods used in clinical practice.
Finally, we present a new hybrid cellular automata model of far greater com-
plexity than those in existing literature incorporating far more detail of the host
immune system. This model is able to describe the complex mechanisms of the im-
mune system and impact of chemokines on the tumour system. The model allows
for a description of each individual cell and its interactions with other cells and
the surrounding chemical fields. The effect of chemokines in a growing tumour and
the simulation of virtual patients using similar Kaplan-Meier curve also provide a
new contribution of the literature.
In summary, in this thesis a number of useful models have been developed that
can be employed as preliminary investigative tools for experimentalists who con-
duct expensive in vitro and in vivo experiments to test and refine hypotheses prior
to entering the lab. With further cross disciplinary collaboration, these type of
models can be refined to provide a more accurate description of the underlying
cancer biology and hence yield more relevant predictions and tests of hypotheses.
135
5.3 Future research
These mathematical models using ordinary differential equations, optimal control
theory and hybrid cellular automata models provide new insights into the effects
of the immune system on a growing tumour. Further development of the ODE
models developed here will include the extension to consider the chemokine secre-
tion involving CD4+ T helper cells. Furthermore, development of PDEs models to
include spatial effects from ODE models will be observed in future research. The
consideration of combined immunotherapy-chemotherapy treatment strategies may
also be analysed using optimal control theory.
Future developments based upon the CA model will involve considering specific
types of cancer. Also, the effect of chemokines on the cell-cell interactions will
be more deeply investigated. More complex partial differential equations related
to chemokines secretion resulting from cell-cell interactions including the tumour
treatment may also be introduced in future work. Finally, 3D CA model also will
be considered in future work.
Bibliography
[1] E.H.J.G. Aarntzen, C.G. Figdor, G.J. Adema, Dendritic cell vaccination
and immune monitoring, Cancer Immunology Immunotherapy, 57:1559-1568,
2008.
[2] J.A. Adam. The dynamic of growth-factor-modified immune response to can-
cer growth: one dimensional model. Mathematical Computational Modelling,
17:83-106, 1993.
[3] J.A. Adam, N. Bellomo. A survey of models for tumor-immune system dy-
namics. Basic models of tumor immune system interactions-identification,
Analysis and Predictions. Birkhauser, Basel, 1997.
[4] Australian Institute of Health & Welfare, Subjects : Cancer, available at
http://www.aihw.gov.au/cancer/, accessed June 7, 2012.
[5] Australian Institute of Health & Welfare, available at http://www.aihw.
gov.au/publications/can/ca06/ca06-c01.pdf, accessed June 7, 2012.
[6] T. Alarcon, H.M. Byrne and P.K. Maini, A cellular automata model for tu-
mour growth in inhomogeneous environment, Journal of Theoretical Biology,
225:257-274, 2003.
[7] P. Allavena, F. Marchesi and A. Mantovani, The role of chemokines and
their receptors in tumor progression and invasion: Potential new target of
biological therapy, Current Cancer Therapy Reviews, 1:81-92, 2005.
[8] G. Albano, V. Giorno and C. Saturnino. A prey-predator model for immune
response and drug resistance in tumor growth. EUROCAST 2007, LNCS
4739, 171-178, 2007.
[9] J.C. Arciero, T.L. Jackson, D. E. Kirschner. A mathematical model of
tumour-immune evasion and siRNA Treatment. Discrete and Continuous
Dynamical Systems-Series B, 4:39-58, 2004.
136
137
[10] R.P. Araujo, D.L.S. McElwain. A history of the study of solid tumour growth:
The contribution of mathematical modelling. Bulletin of Mathematical Biol-
ogy, 66:1039-1091, 2004.
[11] J. Banchereau and R.M. Steinman, Dendritic cells and the control of immu-
nity, Nature, 392:245-252, 1998.
[12] S. Banerjee, R.P. Sarkar. Delay-induced model for tumour-immune interac-
tion and control of malignant tumour growth. BioSystem, 91:268-288, 2008.
[13] Y. Becker. Dendritic cell activity against primary tumours: An overview, In
Vivo, 7:187-191, 1993.
[14] N. Bellomo and L. Preziosi. Modelling and mathematical problems related
to tumor evolution and interaction with the immune system. Mathematical
and Computer Modelling, 32:413-452, 2000.
[15] R. Bhat, S. Dempe, C. Dinsart and J. Rommelaere. Enhancement of NK cell
antitumor responses using an oncolytic parvovirus. International Journal of
Cancer, 2010.
[16] J.N. De Blattman, P.D. Greenberg. Cancer immunotherapy: A treatment
for the masses. Science, 305:200-205, 2004.
[17] R.J. De Boer, P. Hogeweg, H.F. Dullens, et al.. Macrophage T lymphocyte
interactions in the anti-tumor immune response: a mathematical model. The
Journal of Immunology, 134:2748-2758, 1985.
[18] B. Bradie. A Friendly Introduction to Numerical Analysis. Pearson Prentice
Hall. 2006
[19] S. Bunimovich-Mendrazitsky, E. Shochat and L. Stone. Mathematical model
of BCG immunotherapy in superficial bladder cancer. Bulletin of Mathemat-
ical Biology, 69:1847-1870, 2007.
[20] T. Burden, J. Ernstberger and K.R. Fister. Optimal control applied
to immunotherapy. Discrete and Continuous dynamical Systems-Series B,
4(1):135-146, 2004.
[21] S.K. Burgdorf, A. Fischer, P.S. Myschetzky, et al.. Clinical responses in pa-
tients with advanced colorectal cancer to a dendritic cell based vaccine. On-
cology Report, 20:1305-1311, 2008.
[22] A.C. Burton. Rate of growth of solid tumours as a problem of diffusion.
Growth, 30, 159-176, 1966.
138
[23] A. Cappuccio, M. Elishmereni and Z. Agur. Cancer immunotherapy by
interleukin-21: Potential treatment strategies evaluated in mathematical
model. Cancer Research, 66, 2006.
[24] A. Cappuccio, F. Castiglione and B. Piccoli. Determination of the optimal
therapeutic protocols in cancer immunotherapy. Mathematical Biosciences,
doi:10.1016/j.mbs.2007.02.009, 2007.
[25] F. Castiglione, B. Piccoli. Optimal control in a model of dendritic cell trans-
fection cancer immunotherapy. 43rd IEEE Conference on Decision and Con-
trol, December 14-17, 2004.
[26] F. Castiglione, B. Piccoli. Optimal control in a model of dendritic cell trans-
fection cancer immunotherapy. Bulletin of Mathematical Biology, 68:255-274,
2006.
[27] F. Castiglione, B. Piccoli. Cancer immunotherapy, mathematical modeling
and optimal control. Journal of Theoretical Biology, 247:723-732, 2007.
[28] S. Chareyron, M. Alamir. Mixed immunotherapy and chemotherapy of tu-
mors: Feedback design and model updating schemes. Journal of Theoretical
Biology, 258:444-454, 2009.
[29] A.J. Codman, J.M. Murray. Optimal control for a stochastic model of cancer
chemotherapy. Mathematical Biosciences, 168:187-200, 2000.
[30] M.A. Cooper, T.A. Fehniger, A. Fuchs, et al.. NK cell and DC interactions.
Trends in Immunology, 25:47-52, 2004.
[31] J. Couzin. Select T cells, given space, shrink tumors. Science, 297:444-454,
2002.
[32] A.I. Daud, N. Mirza, B. Lenox, et al.. Phenotypic and functional analysis
of dendritic cells and clinical outcome in patients with high-risk melanoma
treated with adjuvant granulocyte macrophage colony-stimulating factor.
Journal of Clinical Oncology, 26:3235-3241, 2008.
[33] L. de Pillis, K.R. Fister, W. Gu, et al.. Mathematical model creation for
cancer chemo-immunotherapy. Computational and Mathematical Methods in
Medicine, 10:165-184, 2009.
[34] L.G. de Pillis, A.E. Radunskaya. A Mathematical tumour model with im-
mune resistance and drug therapy: an optimal control approach. Journal of
Theoretical Biology, 3:79-100, 2001.
139
[35] de Pillis, A.E. Radunskaya. A mathematical model of immune response to
tumor invasion. Computational Fluid and Solid Mechanics, 1661-1668, 2003.
[36] L.G. de Pillis, A.E. Radunskaya. The dynamics of an optimally controlled
tumor model: a case study. Mathematical and Computer Modelling, 37:1221-
1244, 2003.
[37] L.G. de Pillis, W. Gu, A.E. Radunskaya. Mixed immunotherapy and
chemotherapy of tumours: modeling, applications and biological interpre-
tations. Journal of Theoretical Biology, 238:841-862, 2006.
[38] L.G. de Pillis, D.G. Mallet and A.E. Radunskaya. Spatial Tumor-Immune
Modeling. Computational and Mathematical Methods in Medicine, 7(2-
3):159-176, 2006.
[39] L.G. de Pillis, A.E. Radunskaya and C.L. Wiseman. A validated mathe-
matical model of cell-mediated immune response to tumour growth. Cancer
Research, 65:7950-7958, 2005.
[40] L.G. de Pillis, W. Gu, K.R. Fister, T. Head, et al.. Chemotherapy for tumors:
An analysis of the dynamics and a study of quadratic and linear optimal
controls. Mathematical Bioscience, 209:292-315, 2007.
[41] H.P. de Vladar and J.A. Gonzalez. Dynamic response of cancer under the
influence of immunological activity and therapy. Journal of Theoretical Bi-
ology, 227:335-348, 2004.
[42] M.V. Dhodapkar, K.M. Dhodapkar, A.K. Palucka. Interactions of tumor
cells with dendritic cells: balancing immunity and tolerance. Cell Death and
Differentiation, 15:39-50, 2008.
[43] A. Diefenbach, E.R. Jensen, A.M. Jamieson, D. Raulet. Rael and H60 ligands
of NKG2D receptor stimulate tumor immunity. Nature, 413:165-171, 2001.
[44] A. d’Onofrio. A general framework for modeling tumour-immune system
competition and immunotherapy: mathematical analysis and biomedical in-
ferences. Physica D, 208:220-235, 2005.
[45] A. d’Onofrio. Metamodeling tumor-immune system interaction, tumor eva-
sion and immunotherapy. Mathematical and Computer Modelling, 47:614-
637, 2008.
[46] P. Dua, V. Dua, E.N. Pistikopoulos. Optimal delivery of chemotherapeutic
agents in cancer. Computer and Chemical Engineering, 32:99-107, 2008.
140
[47] W. Duchting and T. Vogelsaenger. Analysis, forecasting and control of three-
dimensional tumor growth and treatment. Journal of Medical Systems, 8:461-
475, 1984.
[48] R. Eftimie and J.L. Bramson. Interactions between the immune system and
cancer: A brief review of non-spatial mathematical models. Bulletin Mathe-
matical Biology, 73:2-32, 2011.
[49] S. Eikenberry, C. Thalhauser, Y. Kuang. Tumor-Immune Interaction, Sur-
gical Treatment, and Cancer Recurrence in a Mathematical Model of
Melanoma, PloS Computational Biology, 5, 2009.
[50] A. El-Gohary. Chaos and optimal control of equilibrium states of tumor
system with drug, Chaos, Soliton and Fractals, 41:425-435, 2009.
[51] G. Ferlazzo, M.L. Tsang, L. Moretta, et al.. Human Dendritic Cells Activate
Resting Natural Killer (NK) Cells and Are Recognized via the NKp30 Recep-
tor by Activated NK Cells. Journal of Experimental Medicine, 195:343-351,
2002.
[52] Jr.S.C. Ferreira, M.L. Martins, M.J. Vilela. Reaction diffusion model for the
growth of avascular tumor. Physical Review E, 65, 021907, 2002.
[53] Jr.S.C. Ferreira, M.L. Martins, M.J. Vilela. Morphology transitions induced
by chemotherapy in carcinomas in situ. Physical Review E, 67, 051914, 2003.
[54] W.H. Fleming and R.W. Rishel. Deterministic and stochastic optimal control,
Springer Verlag, New York, 1975.
[55] K.R. Fister and J.C. Panetta. Optimal control applied to competing
chemotherapeutic cell-kill strategies. SIAM Journal Applied Mathematics,
63(6):1954-1971, 2003.
[56] K.R. Fister and J.C. Panetta. Optimal control applied to cell-cycle-specific
cancer chemotherapy. SIAM Journal Applied Mathematics, 60(3):1059-1072,
2000.
[57] J. Folkman and M. Hochberg. Self regulation of growth in three dimensions.
Journal of Experimental Medicine, 138:745-753, 1973.
[58] L. Fong and E.G. Engleman. Dendritic cells in cancer immunotherapy. An-
nual Review Immunology, 18:245-273, 2000.
[59] K.A. Foon, J. Yannelli, and M. Bhattacharya-Chatterjee. Colorectal cancer
as a model for immunotherapy. Clinical Cancer Research, 5:225-236, 1999.
141
[60] A. Ghaffari and N. Naserifar. Mathematical modeling and lyapunov-based
drug administration in cancer chemotherapy. Iranian Journal of Electrical
and Electronic Engineering, 5(3):151-158, 2009.
[61] A. Ghaffari and N. Naserifar. Optimal therapeutic protocols in cancer im-
munotherapy. Computers in Biology and Medicine, 40:261-270, 2010.
[62] D. Grecu, A.S. Carstea, A.T. Grecu, et al.. Mathematical modelling of tumor
growth. Romanian Reports in Physics, 59:447-455, 2007.
[63] H.P. Greenspan. Model for the growth of solid tumour by diffusion. Studies
in applied Mathematics, LI, 4:317-340, 1972.
[64] H.P. Greenspan. On the growth and stability of cell cultures and solid tu-
mour, Journal of Theoretical Biology, 56:229-242, 1976.
[65] D.N. Hart. Dendritic cells: Unique leukocyte populations which control the
primary immune response. Blood, 90:3245-3287, 1997.
[66] F.J. Hsu, C. Benike, F. Fagnoni, et al.. Vaccination of patients with B-cell
lymphoma using autologous antigen-pulsed dendritic cells. Nature Medicine,
2:52-58, 1996.
[67] O.G. Isaeva and V.A. Osipov. Modelling of anti-tumour immune response:
Immunocorrective effect of weak centimetre electromagnetic waves. Compu-
tational and Mathematical Methods in Medicine, 10:185-201, 2009.
[68] M. Itik, M.U. Samci, S.P. Branks. Optimal control of drug therapy in cancer
treatment. Nonlinear Analysis, 71:e1473-e1486, 2009.
[69] P.S. Joshi, J.Q. Liu, Y. Wang, et al.. Cytokine-induced killer T cells kill im-
mature dendritic cells by TCR-independent and perporin-dependent mecha-
nisms, Journal of leukocyte Biology, 80:1345-1353, 2006.
[70] B. Joshi, X. Wang, S. Banerjee, et al.. On immunotherapies and cancer
vaccination protocols: A mathematical modelling approach. Journal of The-
oretical Biology, 289:820-827, 2009.
[71] V. Jurisitc, G. Konevic, O. Markovic, et al.. Clinical stage-depending de-
crease of NK cell activity in multiple myeloma patients. Leukemia Research,
31, Supplement 2, 2007.
[72] Y. Kawarada, R. Ganss, N. Garbi, et al.. NK- and CD8+ T cells-mediate
eradication of established tumors by peritumoral injection of CpG-containing
oligodeoxynucleotides. Journal of Immunology, 167(1):5247-5253,2001.
142
[73] T.J. Kindt, R.A. Goldsby, B.A. Osborne. Kuby Immunology. W.H. Freeman
and Company. New York, 2007.
[74] K.L. Kiran, D. Jayachandran and S. Lakshminarayanan. Mathematical mod-
elling of avascular tumour growth based on diffusion of nutrients and its val-
idation. The Canadian Journal of Chemical Engineering, 87:732-740, 2009.
[75] N.F. Kirkby, N.G. Burnet and D.B.F. Faraday. Mathematical modelling of
the response of tumour cells to radiotherapy. Nuclear Instruments and Meth-
ods in Physics Research B, 188:210-215, 2002.
[76] D. Kirschner. On the global dynamics of a model for tumor immunotherapy.
Mathematical Biosciences and Engineering, 6:573-583, 2009.
[77] D. Kirschner and J.C. Panetta. Modeling immunotherapy of the tumor-
immune interaction. Journal of Mathematical Biology, 37:235-252, 1998.
[78] M. Kolev, 2003. Mathematical modelling of the competition between tumors
and immune system considering the role of the antibodies. Mathematical and
Computer Modelling, 37:1143-1152, 2003.
[79] V.A. Kuznetsov, I.A. Makalkin, M.A.Taylor, et al.. Nonlinear dynamics of
immunogenic tumors: Parameter estimation and global bifurcation analysis.
Bulletin of Mathematical Biology, 56:295-321, 1994.
[80] V.A. Kuznetsov, G.D. Knott. Modeling tumour regrowth and immunother-
apy. Mathematical and Computer Modeling, 33:1275-1286, 2001.
[81] N. Larmonier, J. Fraszack, D. Lakomy, et al.. Killer dendritic cells and their
potential for cancer immunotherapy. Cancer Immunology, Immunotherapy,
59:1-11, 2010.
[82] S. Lenhart and J.T. Workman. Optimal control applied to biological models.
Mathematical and Computational Biology Series. Chapman & Hall/CRC.
2007.
[83] A. Lin. A model of tumour and lymphocyte interactions. Discrete and Con-
tinuous Dynamical System - Series B, 4(1):241-266, 2004.
[84] A. Lin, A. Schildknecht, L.T. Nguyen, et al.. Dendritic cells integrate signals
from the tumor microenvironment to modulate immunity and tumor growth.
Immunology Letter, 127:77-84, 2010.
143
[85] P. Lollini, A. Palladini, F. Pappalardo, et al.. Predictive Models in tumour
immunology, N. Bellomo et al. (eds), Selected Topics in cancer Modeling.
Springer Science+Business Media, LLC, 2008.
[86] B. Ludewig, P. Krebs, T. Junt, et al.. Determining control parameters for
dendritic cell-cytotoxic T lymphocyte interaction. European Journal of Im-
munology, 34:2407-2418, 2004.
[87] D.L. Lukes. Differential Equations: Classical to Controlled, Mathematics in
Science and Engineering, Academic Press, New York, 1982
[88] D.G. Mallet. Mathematical modelling of the role of haptotaxis in tumour
growth and invasion, PhD thesis. Queensland University of Technology, Bris-
bane, Australia, 2004.
[89] D.G. Mallet and L.G. de Pillis, 2006. A cellular automata model of tumour-
immune system interactions. Journal of Theoretical Biology, 239:334-350,
2006.
[90] D.G. Mallet and G.J. Pettet. A mathematical model of integrin-mediated
haptotactic cell migration. Bulletin of Mathematical Biology, 68:231-253,
2006.
[91] R.B. Martin. Optimal control drug scheduling of cancer chemotherapy. Au-
tomatica, 28(6):1113-1123, 1992.
[92] A. Matzavinos, M.A.J. Chaplain and V.A. Kuznetsov. Mathematical mod-
elling of the spatio temporal response of cytotoxic T-lymphocytes to a solid
tumour. Mathematical Medicine and Biology, 21:1-34, 2004.
[93] G. Mazzolini, O. Murillo, C. Atorrasagasti, et al.. Immunotherapy and immu-
noescape in colorectal cancer, World Journal Gastroenterology, 13(14):5822-
5831, 2007.
[94] J.M. Murray. Some optimal control problems in cancer chemotherapy with
a toxicity limit. Mathematical Biosciences, 100:49-67, 1990.
[95] T.T. Murooka, S.E. Ward, and E.N. Fish. Chemokines and Cancer, in: Cy-
tokines and Cancer, Springer, New York, 2005.
[96] A. Moretta. Natural killers and dendritic cells: rendezvous in abused tissues.
Nature Reviews Immunology, 2(12):957-964, 2002.
144
[97] B. Mukhopadhyay and R. Bhattacharyya. Temporal and spatiotemporal vari-
ations in a mathematical model of macrophage-tumor interaction. Nonlinear
Analysis: Hybrid Systems, 2:819-831, 2008.
[98] B. Mukhopadhyay and R. Bhattacharyya. A nonlinear mathematical model
of virus-tumour-immune system interaction: Deterministic and stochastic
analysis. Stochastic Analysis and Applications, 27:409-429, 2009.
[99] F. Nani and H.I. Freedman. A mathematical model of cancer treatment by
immunotherapy. Mathematical Biosciences, 163:159-199, 2000.
[100] D. Nagorsen, S. Voigt, E. Berg, et al.. Tumor-infiltrating macrophages and
dendritic cells in human colorectal cancer: relation to local regulatory T
cells, systemic T-cell response against tumor-associated antigens and sur-
vival. Journal of Translational Medicine, 5(62), 2007.
[101] Y. Naito, K. Saito, K. Shiiba, et al.. CD8+ T cells infiltrated within cancer
cell nests as a prognostic factor in human colorectal cancer. Cancer Research,
58:3491-3494, 1998.
[102] R.P. Negus, G.W. Stamp, J. Hadley, et al.. Quantitative assessment of leuko-
cyte infiltrate in ovarium cancer and its relationship to the expression of C-C
chemokines. The American Journal of Pathology, 150:1723-1734, 1997.
[103] A.S. Novozhilov, F.S. Berezovskaya, E.V. Koonin, et al.. Mathematical mod-
eling of tumour therapy with oncolytic viruses: Regimes with complete tu-
mour elimination within the framework of deterministic models. Biology Di-
rect, 1:6, available at http://www.biology-direct.com/content/1/1/6, 2006.
[104] M.R. Owen, H.M. Byrne, C.E. Lewis. Mathematical modelling of the use of
macrophages as vehicles for drug delivery to hypoxic tumour sites. Journal
of Theoretical Biology, 226:377-391, 2004.
[105] J.C. Panetta. A mathematical model of periodically pulsed chemotherapy:
Tumor recurrence and metastasis in a competitive environment. Bulletin of
Mathematical Biology, 58(3): 425-447, 1996.
[106] C.R. Parish. Cancer immunotherapy: the past, the present and the future.
Immunology and Cell Biology, 81:106-133, 2003.
[107] A.J. Perumpanani, D.L. Simmons, A.J.H. Gearing, et al.. Extracellular
matrix-mediated chemotaxis can impede cell migration Proceeding of the
Royal Society, London B, 265:2347-2352, 1998.
145
[108] A.J. Perumpanani, J.A. Sherratt, J. Norbury, et al.. Biological inferences
from a mathematical model for malignant invasion. Invasion Metastasis,
16:209-221, 1996.
[109] A.J. Perumpanani, J.A. Sherratt, J. Norbury, et al.. A two parameter family
of traveling waves with a singular barrier arising from the modelling of ex-
tracellular matrix mediated cellular invasion. Physica D, 126:145-159, 1999.
[110] G.J. Pettet, C.P. Please, M.J. Tindall, et al.. The migration of cells in mul-
ticell tumour speroid. Bulletin of Mathematical Biology, 63:231-257, 2001.
[111] L.S. Pontryagin, V.G. Boltyanskii, R.V. Gamkrelidze, et al.. The Mathemat-
ical Theory of Optimal Control Processes, Vol 4, Gordon and Breach Science
Publishers, 1986.
[112] L. Preziosi. Cancer modelling and simulation. A CRC Press Company, UK,
2003.
[113] L. Preziosi, A. Tosin. Multiphase modelling of tumour growth and extra-
cellular matrix interaction: mathematical tools and applications. Journal of
Mathematical Biology, 58:625-656, 2009.
[114] D. Raman, P.J. Baugher, Y.M. Thu., et al.. Role of chemokines in tumour
growth. Cancer Letter, 256:137-165, 2007.
[115] B. Ribba, T. Alarcon, K. Marron, et al.. The use of hybrid cellular automata
models for improving cancer therapy. ACRI 2004, LNCS 3305, 444-453, 2004.
[116] H. Riedel. Models for tumour growth and differentiations. In M.R. Alison
(ed), The Cancer Handbook. Wiley, 2004.
[117] R.W. Ruddon. Cancer Biology, Oxford University Press, 2007.
[118] M.L. Salem and D.J. Cole. Dendritic cell recovery post-lymphodepletion: a
potential mechanism for anti-cancer adaptive T cell therapy. Cancer Imm-
nunology, Immunotherapy, 59:341-353, 2010.
[119] M.H. Sandel, et al.. Prognostic Value of Tumor-Infiltrating Dendritic Cells
in Colorectal Cancer: Role of Maturation Status and Intra-tumoral Local-
ization. Clinical Cancer Research, 11(7):2576-2582, 2005.
[120] G. Schuler, B. Schuler-Thurner, R.M. Steiman, The use of dendritic cells in
cancer immunotherapy. Current Opinion in Immunology, 15:138-147, 2003.
[121] C.L. Smith, N. Dylphy, M. Salio, and V. Cerundolo, Immunotherapy of col-
orectal cancer, British Medical Bulletin, 2002, 64:181-200.
146
[122] M. Smyth, D. Godfrey, J. Trapani. A fresh look at tumor immunosurveillance
and immunotherapy. Nature Immunology, 2:293-299, 2001.
[123] R.M. Steinman. The dendritic cell system and its role in immunogenicity.
Annual Review of Immunology, 9:271-296, 1991.
[124] R. Somasundaram, D. Herlyn. Chemokines and the microenvironment in neu-
roectodermal tumor-host interaction. Seminars in Cancers Biology, 19:92-96,
2009.
[125] G.W. Swan. Role of optimal control theory in cancer chemotherapy. Mathe-
matical Biosciences, 101:237-284, 1990.
[126] A. Swierniak, U. Ledzewicz, H. Schattler. Optimal control for a class of com-
partmental models in cancer chemotherapy. International Journal of Applied
Mathematics and Computer Science, 13(3):357-368, 2003.
[127] A.P. Vicari, S. Ait-Yahia, K. Chemin, et al.. Antitumor effects of the mouse
chemokine 6Ckine/SLC through angiostatic and immunological mechanisms.
The Journal of Immunology, 165:1992-2000, 2000.
[128] J.P. Ward and J.R. King. Mathematical modelling of a vascular-tumour
growth. IMA Journal of Mathematics Applied in Medicine & Biology, 14:39-
69, 1997.
[129] S. Wilson and D. Levy. A mathematical model of the enhancement of tu-
mor vaccine efficacy by immunotherapy. Bulletin of Mathematical Biology,
74:1485-1500, 2012.
[130] J.T. Wu, D.H. Kirn, L.M. Wein. Analysis of a three-way race between tu-
mour growth, a replication-component virus an immune response. Bulletin
of Mathematical Biology, 66:605-625, 2004.
[131] World Health Organization, Media Centre, Fact sheets: Cancer, avail-
able at http://www.who.int/mediacentre/factsheets/fs297/en/, ac-
cessed June, 2012.
[132] Y. Wu, L. Xia, M. Zhang, X. Zhao. Immunodominance analysis through
interactions of CD8+T cells and DCs in lymph nodes. Mathematical Bio-
sciences, 2010.